This feed does not validate.
In addition, interoperability with the widest range of feed readers could be improved by implementing the following recommendations.
line 33, column 0: (16 occurrences) [help]
<site xmlns="com-wordpress:feed-additions:1">137950579</site> <item>
line 105, column 0: (19 occurrences) [help]
line 225, column 0: (34 occurrences) [help]
line 320, column 0: (4 occurrences) [help]
line 320, column 0: (4 occurrences) [help]
line 505, column 0: (17 occurrences) [help]
line 697, column 3: (55 occurrences) [help]
]]></content:encoded>
^
line 948, column 0: (6 occurrences) [help]
line 1180, column 0: (24 occurrences) [help]
line 1294, column 0: (4 occurrences) [help]
<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:wfw="http://wellformedweb.org/CommentAPI/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:sy="http://purl.org/rss/1.0/modules/syndication/" xmlns:slash="http://purl.org/rss/1.0/modules/slash/" xmlns:georss="http://www.georss.org/georss" xmlns:geo="http://www.w3.org/2003/01/geo/wgs84_pos#" > <channel> <title>Kavita Ganesan, PhD </title> <atom:link href="https://kavita-ganesan.com/feed/" rel="self" type="application/rss+xml" /> <link>https://kavita-ganesan.com/</link> <description></description> <lastBuildDate>Wed, 05 Apr 2023 23:02:05 +0000</lastBuildDate> <language>en-US</language> <sy:updatePeriod> hourly </sy:updatePeriod> <sy:updateFrequency> 1 </sy:updateFrequency> <generator>https://wordpress.org/?v=6.5.2</generator> <image> <url>https://kavita-ganesan.com/wp-content/uploads/cropped-KavitaOutdoor1-V2-1-edited-1-scaled-1-32x32.jpeg</url> <title>Kavita Ganesan, PhD </title> <link>https://kavita-ganesan.com/</link> <width>32</width> <height>32</height></image> <site xmlns="com-wordpress:feed-additions:1">137950579</site> <item> <title>Exploring the Ethical Implications of AI: A Closer Look at the Challenges Ahead</title> <link>https://kavita-ganesan.com/ai-ethical-issues-and-examples/</link> <comments>https://kavita-ganesan.com/ai-ethical-issues-and-examples/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Wed, 29 Mar 2023 20:16:20 +0000</pubDate> <category><![CDATA[AI ETHICS]]></category> <category><![CDATA[AI For Business]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8759</guid> <description><![CDATA[This article discusses five common ethical issues that arise when AI is not implemented or released in the most responsible way.]]></description> <content:encoded><![CDATA[<p>AI ethics is about releasing and implementing AI responsibly, paying attention to several considerations, from <em>data etiquette</em> to<em> tool development risks, </em>as discussed in a <a href="https://www.opinosis-analytics.com/blog/what-is-ai-ethics/">previous</a> article. In this article, we’ll explore some of the ethical issues that arise with AI systems, particularly machine learning systems, when we overlook the ethical considerations of AI, often unintentionally.</p> <p></p> <h2 class="wp-block-heading">The 5 Common AI Ethical Issues </h2> <p></p> <h3 class="wp-block-heading" id="ai-bias">1. Bias propagation </h3> <p>Although there’s a strong belief that algorithms are less biased than humans, AI systems are known to propagate our conscious and unconscious biases. </p> <p>For example, there are known recruiting tools that algorithmically “learned” to <a href="https://www.reuters.com/article/us-amazon-com-jobs-automation-insight/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G">dismiss women candidates</a> as they learned that men were preferred in the tech workforce. </p> <p>Even facial recognition systems are infamous for disproportionately making mistakes on <a href="https://news.mit.edu/2018/study-finds-gender-skin-type-bias-artificial-intelligence-systems-0212">minority groups and people of color</a>. For example, when the researcher, Joy Buolamwini, looked into the accuracy of facial recognition systems from various companies, she found that the error rate for lighter-skinned males was no higher than 1%. However, for darker-skinned females, the mistakes were much more significant, reaching up to 35%. Even the most renowned AI systems have been unable to accurately identify female celebrities of color.</p> <p>So, what’s the primary cause of AI bias?</p> <p>Data. AI systems today are only as good as the data they are trained on; if the data is nonrepresentative, skewed towards a particular group, or somehow imbalanced, the AI system will learn this nonrepresentation and propagate biases. </p> <p>Bias in data can be caused by a range of factors. For example, if historically, certain groups of people have been discriminated against, this discrimination will be very well recorded in the data. </p> <p>Another reason for bias in data can be a company’s data warehousing processes or lack thereof, causing AI systems to learn from skewed samples of data instead of representative ones. Even using a snapshot of the Web to train models can mean you’ve learned the biases in that snapshot. This is why large language models (LLMs) are not free from biases when they’re quizzed on subjective topics. </p> <p>Bias in data can also be a <em>development mistake </em>where the data used for model development was not sampled correctly, resulting in an imbalance of subgroup samples. </p> <p></p> <p></p> <div class="wp-block-stackable-columns stk-block-columns stk-block stk-a43493a" data-block-id="a43493a"><div class="stk-row stk-inner-blocks stk-block-content stk-content-align stk-a43493a-column"><div class="wp-block-stackable-column stk-block-column stk-block-column--v3 stk-column stk-block-column--v2 stk-block stk-13a5603" data-block-id="13a5603"><div class="stk-column-wrapper stk-block-column__content stk-container stk-13a5603-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-13a5603-inner-blocks"><p class="has-ast-global-color-4-background-color has-background"><strong>Bottom line:</strong> When there’s limited oversight of the quality of data used for <a href="https://www.opinosis-analytics.com/ai-glossary/machine-learning-model-training">model training</a>, various unintended biases are bound to happen. We may not know when and where especially with unconstrained multi-taskers like LLMs. </p></div></div></div></div></div> <h3 class="wp-block-heading">2. Unintended Plagiarism</h3> <p>Generative AI tools such as <a href="https://www.opinosis-analytics.com/blog/what-is-gpt-3-and-what-it-can-do-for-your-business">GPT-3 </a>and ChatGPT learn from massive amounts of Web data. These tools generate the probability of producing meaningful content. In doing that, these generative AI tools may repeat content on the Web word-for-word without any attribution. </p> <p>How would we know that the generated content is, in fact, unique? What if the uniquely generated text is identical to a source on the Web? Can the source claim plagiarism?</p> <p>We’re already seeing this issue in artwork generators that learn from a large number of art pieces belonging to different artists. The AI tool may end up generating art that combines work from multiple artists. </p> <p>In the end, who exactly owns the copyright to the generated art? If the artwork is too similar to existing ones, this can lead to copyright infringement.</p> <div class="wp-block-stackable-columns stk-block-columns stk-block stk-92b89b8" data-block-id="92b89b8"><div class="stk-row stk-inner-blocks stk-block-content stk-content-align stk-92b89b8-column"><div class="wp-block-stackable-column stk-block-column stk-block-column--v3 stk-column stk-block-column--v2 stk-block stk-db68522" data-block-id="db68522"><div class="stk-column-wrapper stk-block-column__content stk-container stk-db68522-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-db68522-inner-blocks"><p class="has-ast-global-color-4-background-color has-background"><strong>Bottom line:</strong> Leveraging Web and public datasets for developing models can result in unintended plagiarism. However, due to little AI regulation worldwide, we currently lack enforceable solutions.</p></div></div></div></div></div> <h3 class="wp-block-heading">3. Technology Misuse</h3> <p>A while ago, a Ukrainian state leader was<a href="https://www.newsweek.com/fact-check-russia-ukraine-deepfake-technology-video-call-1752690"> portrayed as saying something they did not </a>actually say, using a tool called <em><a href="https://en.wikipedia.org/wiki/Deepfake">deepfakes</a></em>. This AI tool can generate videos or images of people saying things that they never actually said. Similarly, AI image generator tools like DALL.E and Stable Diffusion can be used to create incredibly realistic depictions of events that never occurred.</p> <p>Intelligent tools like these can be used as weapons in a war (as we’ve already seen), to spread misinformation to gain political advantage, manipulate public opinion, commit fraud, and more. </p> <p>In all of these, AI is <em>NOT</em> the bad actor, it’s doing what it’s designed to do. The bad actors are the humans who misuse AI for their own advantage. Furthermore, the companies or teams that create and distribute these AI tools have not taken into account the wider effects these tools may have on society, which is also an issue. </p> <div class="wp-block-stackable-columns stk-block-columns stk-block stk-2aa8b47" data-block-id="2aa8b47"><div class="stk-row stk-inner-blocks stk-block-content stk-content-align stk-2aa8b47-column"><div class="wp-block-stackable-column stk-block-column stk-block-column--v3 stk-column stk-block-column--v2 stk-block stk-f950917" data-block-id="f950917"><div class="stk-column-wrapper stk-block-column__content stk-container stk-f950917-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-f950917-inner-blocks"><p class="has-ast-global-color-4-background-color has-background"><strong>Bottom line: </strong>While the misuse of technology is not exclusive to AI, because AI tools are so adept at replicating human abilities, it is possible that the abuse of AI could go undetected and have a lasting effect on our view of the world.</p></div></div></div></div></div> <h3 class="wp-block-heading">4. Uneven Playing Fields</h3> <p>Algorithms can be easily tricked, and the same is true of AI-powered software, where you can trick the underlying algorithms to gain an unfair advantage. </p> <p>In a <a href="https://www.linkedin.com/posts/kavita-ganesan_looking-for-a-new-job-new-rules-may-save-activity-7032891511004434432-WKfg?utm_source=share&utm_medium=member_desktop">LinkedIn post</a> that I put out, I discussed how people might trick AI hiring tools when you disclose the attributes the system will use in the decision-making process. </p> <p>While enforcing steps <a href="https://news.yahoo.com/looking-for-a-new-job-new-rules-may-save-you-from-the-ai-hiring-bots-121308792.html">to reveal an AI’s decision-making process in hiring</a> is a well-intentioned step toward promoting transparency, it may enable people to game the system. For example, candidates may learn that certain keywords are preferred in the hiring process and stuff their resumes with such keywords, unfairly getting ranked higher than more qualified candidates. </p> <p>We see this on a much bigger scale with the SEO industry, estimated to be worth over 60 billion dollars. Getting ranked highly in Google’s eyes these days is not just a function of having meaningful content worth reading. But also a function of having done “good SEO” and thus, the growing popularity of this industry. </p> <p>SEO services have enabled organizations with hefty budgets to dominate the ranks as they’re able to invest heavily in creating massive amounts of content, performing keyword optimization, and getting links placed broadly around the Web. </p> <p>While some SEO practices are mere content optimization, some “trick” the search algorithms into believing that their websites are the best in class, the most authoritative, and will provide the best value to readers. This may or may not be true. The highly ranked companies may have just <em>invested in more SEO</em>. </p> <div class="wp-block-stackable-columns stk-block-columns stk-block stk-c5ea403" data-block-id="c5ea403"><div class="stk-row stk-inner-blocks stk-block-content stk-content-align stk-c5ea403-column"><div class="wp-block-stackable-column stk-block-column stk-block-column--v3 stk-column stk-block-column--v2 stk-block stk-808c87c" data-block-id="808c87c"><div class="stk-column-wrapper stk-block-column__content stk-container stk-808c87c-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-808c87c-inner-blocks"><p class="has-ast-global-color-4-background-color has-background"><strong>Bottom line: </strong>Gaming AI algorithms is one of the easiest ways to gain an unfair advantage in business, career, influencer-ship, and politics. People who figure out how your algorithm “operates” and makes decisions can abuse and game the system. </p></div></div></div></div></div> <h3 class="wp-block-heading">5. Widespread Misinformation</h3> <p>As we rely more and more on answers and content generated by generative AI systems, the “facts” that these systems produce can be assumed to be the ultimate truth. For example, in Google’s demo of their generative AI system, Bard, it provides three points in response to the question, <em>“What new discoveries from the James Webb Space Telescope can I tell my 9-year-old about?”</em> One of the points states that the telescope “took the very first pictures of a planet outside of our own solar system.” However, astronomers later pointed out in a very public way that this <a href="https://www.theverge.com/2023/2/8/23590864/google-ai-chatbot-bard-mistake-error-exoplanet-demo">wasn’t the case</a>. Directly using output from such systems can result in widespread misinformation. </p> <p>Unfortunately, without proper citation, it isn’t easy to verify facts and decide which answers to trust and which not to. And as more people accept the content generated without question, this can lead to the spread of false information on a much larger scale than seen with traditional search engines. </p> <p>The same is true for content ghostwritten by generative AI systems. Previously, human ghostwriters had to research information from trustworthy sources, piece them together in a meaningful way, and cite the sources before they publish. But now, they can have entire articles ghostwritten for them by an AI system. Unfortunately, if an article generated by an AI system is published without further verification of the facts, misinformation is bound to spread. </p> <div class="wp-block-stackable-columns stk-block-columns stk-block stk-ca8f877" data-block-id="ca8f877"><div class="stk-row stk-inner-blocks stk-block-content stk-content-align stk-ca8f877-column"><div class="wp-block-stackable-column stk-block-column stk-block-column--v3 stk-column stk-block-column--v2 stk-block stk-0e530db" data-block-id="0e530db"><div class="stk-column-wrapper stk-block-column__content stk-container stk-0e530db-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-0e530db-inner-blocks"><p class="has-ast-global-color-4-background-color has-background"><strong>Bottom line:</strong> Over-reliance on AI-generated content without the human verification element of the facts will have a lasting impact on our worldviews due to the non-fact-checked information we consume over extended periods of time.</p></div></div></div></div></div> <figure class="wp-block-image size-large"><img decoding="async" src="https://www.opinosis-analytics.com/wp-content/uploads/2023/02/5-ethical-implications-of-AI-1024x1024.png" alt="" class="wp-image-17394"/></figure> <h2 class="wp-block-heading">Summary</h2> <p>In this article, we explored some potential ethical issues that can arise from AI systems, particularly machine learning systems. We discussed how:</p> <ul><li> AI systems can propagate racial, gender, age, and socioeconomic biases</li> <li>AI can infringe on copyright laws</li> <li>AI can be used in nonethical ways to harm others</li> <li>AI can be tricked, unleveling the playing field for people and businesses </li> <li>Trusting answers blindly from AI systems can cause widespread misinformation</li></ul> <p>It’s critical to note that many of these problems were <em>not intentionally</em> created, but rather they are the side effects of how these systems were developed, disseminated, and used in practice. </p> <p></p> <p>Although we can’t eliminate these ethical problems entirely, we can certainly take steps in the right direction to minimize the issues created by technology in general, and in this case, AI.</p> <p></p> <p>With insights into the ethical dilemmas of AI, let’s focus on devising strategies for more responsible development and dissemination of AI systems. Instead of waiting for government regulation, in an upcoming article, we’ll explore how businesses can lead the way in doing AI responsibly. </p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/ai-ethical-issues-and-examples/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8759</post-id> </item> <item> <title>What is An AI Strategy and Why Every Business Needs One</title> <link>https://kavita-ganesan.com/what-is-an-ai-strategy-and-why-every-business-needs-one/</link> <comments>https://kavita-ganesan.com/what-is-an-ai-strategy-and-why-every-business-needs-one/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Tue, 14 Mar 2023 20:24:05 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Strategy]]></category> <category><![CDATA[undefined]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8937</guid> <description><![CDATA[This article discusses what an AI strategy means, the different types of AI strategies that you should know about, and how as a leader you can get started with an AI strategy. What Is An AI strategy? An AI strategy may seem like a complicated business-speak, but it’s simply a vision or high-level plan for …<p class="read-more"> <a class="" href="https://kavita-ganesan.com/what-is-an-ai-strategy-and-why-every-business-needs-one/"> <span class="screen-reader-text">What is An AI Strategy and Why Every Business Needs One</span> Read More »</a></p>]]></description> <content:encoded><![CDATA[<p>This article discusses what an AI strategy means, the different types of AI strategies that you should know about, and how as a leader you can get started with an AI strategy.</p> <div class="wp-block-uagb-table-of-contents uagb-toc__align-left uagb-toc__columns-1 uagb-block-e321995f " data-scroll= "1" data-offset= "30" style="" > <div class="uagb-toc__wrap"> <div class="uagb-toc__title"> Table Of Contents </div> <div class="uagb-toc__list-wrap"> <ol class="uagb-toc__list"><li class="uagb-toc__list"><a href="#what-is-an-ai-strategy" class="uagb-toc-link__trigger">What Is An AI strategy?</a><li class="uagb-toc__list"><a href="#why-do-businesses-need-an-ai-strategy" class="uagb-toc-link__trigger">Why Do Businesses Need an AI Strategy?</a><ul class="uagb-toc__list"><li class="uagb-toc__list"><a href="#product-level-ai-strategy" class="uagb-toc-link__trigger">Product Level AI Strategy</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#business-unit-level-ai-strategy" class="uagb-toc-link__trigger">Business Unit Level AI Strategy</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#organizational-level-ai-level-strategy" class="uagb-toc-link__trigger">Organizational Level AI Level Strategy</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#ai-startup-strategy" class="uagb-toc-link__trigger">AI Startup Strategy</a></li></ul></li><li class="uagb-toc__list"><a href="#getting-started-with-your-ai-strategy" class="uagb-toc-link__trigger">Getting Started With Your AI Strategy</a><li class="uagb-toc__list"><a href="#keep-learning-succeed-with-ai" class="uagb-toc-link__trigger">Keep Learning & Succeed With AI</a></ul></ol> </div> </div> </div> <p></p> <p></p> <h2 class="wp-block-heading" id="what-is-an-ai-strategy">What Is An AI strategy?</h2> <p>An AI strategy may seem like a complicated business-speak, but it’s simply a vision or high-level plan for integrating AI into the organization, such that it aligns with your broader business and automation goals. This high-level plan can be a:</p> <ul><li>Product-level AI strategy (I often call this AI product strategy)</li> <li>Business-unit-level AI strategy</li> <li>Organizational-level AI strategy</li> <li>AI startup strategy</li></ul> <p>The granularity of the plan is often inversely proportional to the magnitude of the vision. So, the bigger the vision, the broader the objectives until each objective is broken down into a roadmap for implementation. </p> <h2 class="wp-block-heading" id="why-do-businesses-need-an-ai-strategy">Why Do Businesses Need an AI Strategy?</h2> <p>So now the question becomes, why does an organization need an AI strategy? Why can’t you just dive into implementation or make an off-the-shelf purchase with little planning? </p> <p>As Dale Carnegie says, “An hour of planning can save you 10 hours of doing” and with AI it can save you much more than that. It can save you months in setbacks and unnecessary spending. Let’s look at how an AI strategy can be helpful. </p> <h3 class="wp-block-heading" id="ai-product-strategy">Product Level AI Strategy</h3> <p>At the product level, an AI strategy is needed so that you have the foundational building blocks to ensure that an AI project can progress from an idea to a useful tool to creating meaningful results for the business. </p> <p>The building blocks at this level, could be the data that’s needed to support the initiative, the personnel who will build your data pipelines and develop the relevant models, the metrics to track success, or the competing off-the-shelf solutions that you should consider for integration. Without this level of detailed planning, you’re bound to run into roadblocks that will keep setting you back months at a time. </p> <figure class="wp-block-image aligncenter"><img decoding="async" src="https://www.opinosis-analytics.com/wp-content/uploads/2023/03/ai-product-strategy-1024x559.png" alt="" class="wp-image-17536"/><figcaption class="wp-element-caption">Elements of an AI Strategy at the Product Level</figcaption></figure> <p>For example, in early 2020, one of our healthcare customers wanted to develop an AI tool to automatically extract specific information from clinical records, primarily containing free-form text. They had been doing this <a href="https://www.opinosis-analytics.com/blog/nlp-applications/#7-information-extraction">information extraction</a> work manually for years to obtain reimbursement from insurance companies. </p> <p>But as we were digging deeper into the project, shockingly, we discovered that none of that data was available. What happened?</p> <p>As part of the company’s manual process, they copied data directly from the clinical records and pasted them into third-party forms for reimbursement. Unfortunately, once the forms were submitted, the company kept no records of the information that they copied over. This meant—no historical data for AI. As a result, the company had to first augment its manual processes with <a href="https://www.opinosis-analytics.com/blog/machine-learning-training-data/">data collection</a> in mind. This set them back six whole months.</p> <p>This is a common problem for many companies, where the data to build the AI product is often nonexistent, is in the wrong format, or is yet to be collected. This is not just data for<a href="https://www.opinosis-analytics.com/ai-glossary/machine-learning-model-training"> model training</a> but also data for evaluation and learning more about the problem that you’re solving.</p> <p>The building blocks for every AI product can look slightly different depending on whether you’re customizing an existing solution, building from scratch, or <a href="https://www.opinosis-analytics.com/blog/buying-ai-solutions/">buying an off-the-shelf </a>solution. </p> <h3 class="wp-block-heading" id="business-unit-ai-strategy">Business Unit Level AI Strategy</h3> <p>At the business unit level, an AI strategy is needed to ensure that you’re pursuing AI opportunities that will give you the biggest improvement in your bottom line, productivity, or decision-making. </p> <p>Some of the drivers of those decisions include the severity of the problems that you’re addressing, data availability, infrastructure readiness, availability of shared resources, and in many cases, you’d have to ensure that the selected initiatives align with the business’s broader objectives. </p> <p>To prevent you from tackling the <em>wrong </em>initiatives first (i.e., low-impact problems, no data problems, no buy-in problems), requires planning, which can help you be very strategic in implementing AI in your business unit. </p> <p>For example, if a new problem has no available data for development and evaluation, you may want to start the process manually first. And by doing this, you’re helping your future AI initiative in the following ways:</p> <ol><li>Establish a baseline performance, which you can later use for A/B testing</li> <li>Generate high-quality data</li> <li>Get a better handle on the problem</li> <li>Leverage domain expertise in developing your AI solution in the future </li></ol> <p>To sum up, a business unit level AI strategy is primarily to create a logical AI implementation roadmap to maximize impact.</p> <h3 class="wp-block-heading" id="organizational-ai-strategy">Organizational Level AI Level Strategy</h3> <p>An organizational-level AI strategy is all about getting your organization as a whole <em>ready </em>and <em>prepared </em>for AI. This ensures that you’re not just running one-off pilots in an isolated business unit, but teams across the organization have AI on their radar, know how to tap into the necessary resources to get AI projects going, know how to spot AI opportunities, and have a base level of AI understanding. </p> <p>Getting an organizational-level AI strategy up is more than just finding AI opportunities to go after in your organization. It’s also about identifying all the readiness gaps from the perspective of:</p> <ul><li>Budget (B)</li> <li>Culture (C)</li> <li>Infrastructure (I)</li> <li>Data (D)</li> <li>Skills (S)</li></ul> <p>All of this is so that a company can take an AI idea from conception to implementation to reaping measurable benefits—all with minimal friction, repeatedly.</p> <p>In <a href="https://www.opinosis-analytics.com/aibusinesscasebook">The Business Case For AI</a>, I talk about the five preparation pillars for AI, referred to as B-CIDS (pronounced, “be kids”), which encapsulates the elements we discussed above.</p> <figure class="wp-block-image aligncenter size-large"><img fetchpriority="high" decoding="async" width="1024" height="611" src="https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-1024x611.png" alt="" class="wp-image-8961" srcset="https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-1024x611.png 1024w, https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-300x179.png 300w, https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-150x89.png 150w, https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-768x458.png 768w, https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-1536x916.png 1536w, https://kavita-ganesan.com/wp-content/uploads/bcids_5_ai_preparation_elements-1-2048x1222.png 2048w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption class="wp-element-caption">The 5 AI Preparation Pillars From <a href="http://amazon.aibusinesscasebook.com/">The Business Case For AI</a> for an Organizational-Level AI Strategy Development. Data readiness is the biggest preparation pillar, followed by cultural readiness.</figcaption></figure> <p>Each pillar is critical, with its own set of requirements for building an AI-ready company.</p> <p>For example, under cultural readiness, there are <strong>six elements</strong> to consider to prepare a company culturally for AI. This is due to fears around the technology, the ethical dilemmas, the uncertainties that AI brings, and the need to be highly data-driven. </p> <p>If you’re not culturally ready for AI, you may experience resistance internally to the adoption of the technology, your employees may not know how to collaborate effectively on AI projects, and the risks that come with AI, including mitigation strategies, may not be well understood.</p> <p>Another critical pillar, data readiness, is all about ensuring your data warehousing is strong and that you’re collecting data from the daily running of your business and are digitally ready to leverage your data stores. Essentially, this is all about getting your<a href="https://www.opinosis-analytics.com/blog/big-data-strategy/"> data strategy</a> up and running.</p> <p>Building up each readiness pillar takes time and a whole lot of planning. And a plan alone is not enough, you need an ACTIONABLE plan—something you can implement.</p> <p>To facilitate the planning process, I’ve outlined<strong> The Jumpstart AI Approach</strong> in The Business Case For AI for getting started with your AI strategy using actionable, short-term strides. In a future article, we’ll explore the approach.</p> <h3 class="wp-block-heading" id="ai-startup-strategy">AI Startup Strategy</h3> <p>An AI startup is either selling an AI tool or technique that you can use within your applications (e.g., Perspective API), or AI is a big part of the startup’s products and services (e.g., Jasper.ai, the content writing assistant). </p> <p>If you’re looking to build a startup with AI as the primary driver of your business, then an AI strategy for your product is a must.</p> <p>Depending on whether you’re developing the AI solution from scratch or using an off-the-shelf model to power your applications, the priorities can look slightly different. However, having worked with multiple AI startups, I can say that these are some common concerns:</p> <ul><li><strong>Model/Technology</strong><ul><li>Are you leveraging an open-source solution, buying, or building from scratch?</li> <li>Who will perform the model development or customization, testing and integration?</li> <li>How will the solution integrate into your software or, how will users access the solution?</li></ul></li> <li><strong>Performance</strong><ul><li>How will you assess the quality of your tool in solving a broad spectrum of unseen problems?</li> <li>How do you ensure the quality of your tools stays consistent with time?</li></ul></li> <li><strong>Monetization</strong><ul><li>How will you monetize your tool?</li></ul></li> <li><strong>Safety</strong><ul><li>How do you ensure that your tool produces the expected output?</li> <li>How will your tool fail gracefully instead of annoying users?</li></ul></li> <li><strong>Messaging</strong><ul><li>What exactly are you selling? The technology and quality behind the technology or the product’s benefits to the end user?</li></ul></li> <li><strong>Data</strong><ul><li>How will you start collecting data from day one to improve models, develop new models, etc?</li></ul></li> <li><strong>Adoption</strong><ul><li>How will you ensure that you’re addressing fears around AI and helping new users become open to using your solution?</li></ul></li> <li><strong>Feedback Loop </strong><ul><li>How would you collect feedback from users to improve your offerings continuously?</li></ul></li></ul> <p>As you’ll notice, this has elements of an AI product strategy and some organizational-level AI strategy, with some monetization and marketing strategy. AI startups must prioritize their messaging, monetization strategy, quality consistency, and feedback loop, as AI is at the front and center of their business.</p> <p>This is not to say that these elements are not critical to non startup AI products that augment existing workflows. It’s just that the priorities in those scenarios could look very different. </p> <p>For example, for a large company seeking to enhance a workflow, a 50% boost in productivity through AI may suffice regardless of the model’s accuracy as long as it meets the task requirements. But for an AI startup selling an AI tool, poor or even just “acceptable” accuracy can destroy a buyer’s trust. </p> <figure class="wp-block-image aligncenter"><img decoding="async" src="https://www.opinosis-analytics.com/wp-content/uploads/2023/03/ai-strategy-for-ai-startups.png" alt="" class="wp-image-17594"/><figcaption class="wp-element-caption">AI startup strategy has elements of an AI product strategy and organizational-level AI strategy.</figcaption></figure> <h2 class="wp-block-heading" id="getting-started-with-your-ai-strategy">Getting Started With Your AI Strategy</h2> <p>The first step in getting started with your AI strategy is to understand what it is, why it’s important, and if you’re considering an organizational, business unit, or product-level strategy. The scope of your AI strategy will determine what preparation elements to focus on.</p> <p>At the product level, we saw that we’re primarily focused on the building blocks related to the specific AI project, such as data and personnel. </p> <p>But at the organizational level, we’re focused on getting the entire organization to have AI on their radar and be able to repeatedly take an AI idea into production and deliver value for the business with minimal friction. This takes much deeper and lengthy preparation along different preparation pillars. </p> <p>For an organizational or business unit level AI strategy, you can use The Business Case For AI, for concrete steps and frameworks to develop your AI strategy. For a product level AI strategy, you’ll have to think about all the elements outlined in the first image.</p> <p>For more hands-on help, you can, of course, <a href="/contact-me">contact me</a> to get help developing your AI strategy.</p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p><em>The post <a href="https://www.opinosis-analytics.com/blog/what-is-ai-strategy/" rel="nofollow">What is An AI Strategy and Why Every Business Needs One</a> appeared first on <a href="https://www.opinosis-analytics.com" rel="nofollow">Opinosis Analytics</a>.</em></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/what-is-an-ai-strategy-and-why-every-business-needs-one/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8937</post-id> </item> <item> <title>What is AI Ethics?</title> <link>https://kavita-ganesan.com/what-is-ai-ethics/</link> <comments>https://kavita-ganesan.com/what-is-ai-ethics/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Tue, 28 Feb 2023 17:48:33 +0000</pubDate> <category><![CDATA[AI ETHICS]]></category> <category><![CDATA[AI For Business]]></category> <category><![CDATA[undefined]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8600</guid> <description><![CDATA[This article is part of a series that will explore what AI ethics means, its implications to society, and how businesses can start leading the way by doing AI responsibly while also reaping its benefits. This article focuses on what ethics means in the context of AI.]]></description> <content:encoded><![CDATA[<p>As AI continues to become more prevalent in our lives, it is crucial to consider the ethical implications of its use. Although AI can augment and revolutionize how we live, work, and interact with each other, it can also cause harm if not used or developed correctly.</p> <p>People can be wrongly imprisoned when facial recognition systems fail in law enforcement and the judicial system. People can be killed if self-driving cars fail to correctly see them as pedestrians on the road. Things can go awfully wrong if we fail to think about the implications of how we use and develop these AI-powered tools. </p> <p>This article is part of a series that will explore what AI ethics means, its implications to society, and how businesses can start leading the way by doing AI responsibly while also reaping its benefits. </p> <p>In this article, we’ll focus on what ethics means in the context of AI.</p> <h2 class="wp-block-heading">What is AI Ethics?</h2> <p>Ethics in the context of AI is all about doing AI responsibly, keeping questions as outlined below in consideration:</p> <h3 class="wp-block-heading"><strong>DATA ETIQUETTE</strong></h3> <p>AI systems today are data-dependent; that’s how machine-learning-driven AI systems learn. The underlying data that you use can significantly impact how the model behaves. The question is are we using this data ethically? When thinking about ethics and data, we have to think about these questions:</p> <ul><li><strong>Where is the data you’re looking to use coming from? </strong>Is it data from your company database, the Web, or a public data bank?</li> <li><strong>Has the data been sourced in the most transparent way considering user privacy? </strong>Did users opt-in to having their data used for model development?</li> <li><strong>Is the data representative of the subgroups of interest? </strong></li></ul> <p>How you source your data, combine, and use that data to train models will impact how your models behave downstream. We’ve seen how the <a href="https://www.opinosis-analytics.com/blog/tay-twitter-bot/">underlying data</a> used by algorithms can impact model behaviors.</p> <p>Not to forget, if you’re using third-party models…the same applies. How the third-party vendor sources data, combines and uses that data to train their models impacts YOUR downstream applications. </p> <p> </p> <p>Bottom line: data etiquette applies to both custom-developed and third-party models you’re building on.</p> <h3 class="wp-block-heading"><strong>EXPLAINABILITY </strong></h3> <p>AI explainability is the ability for AI systems to provide reasoning as to why they arrived at a particular decision, prediction, or suggestion. The explainability of AI systems may not be critical for many AI applications, such as email spam filtering, grammar correction, and product recommendation systems. </p> <p>However, in domains such as healthcare, law enforcement, and any domain where a person’s life, livelihood, and safety are at stake, evidence and explainability are crucial for building trust with AI systems.</p> <p>For example, if an AI system predicts that a patient has a high risk of lung cancer, <em>why did it arrive at that prediction</em>? The insights into the AI’s decision-making will help the physician decide if the recommendations are trustworthy. </p> <p>When it comes to AI explainability, you need to ask if your system must be explainable, and if it does, are you able to gain a glimpse into its reasoning? </p> <h3 class="wp-block-heading"><strong>USAGE ETIQUETTE</strong></h3> <p>AI usage etiquette relates to how you’re integrating AI into a workflow. Is it a sole-decision maker, a human assistant, or a second opinion? How you employ AI can make a massive difference in the risks to users and society when AI gets it wrong.</p> <p>For example, when you use AI to sift through emails to filter out spammy ones automatically, it’s the sole decision maker. Allowing AI to be the sole decision maker in this scenario can be considered low-risk if it makes a wrong decision. Spam may end up in your inbox, or valid emails may get filtered out as spam. Nevertheless, you can still label specific emails in your inbox as spam or browse through your spam emails.</p> <p>However, when you take an application area such as medical diagnosis and treatment planning, asking the AI system to <em>solely </em>make decisions about a cancer treatment plan for a patient is a HUGE risk. Who is to blame if the treatment plan is <em>ineffective </em>by solely trusting the AI tool? The physician? The AI tool or the hospital that decided to employ AI in the first place? </p> <p>Further, usage etiquette is also a function of model accuracy. Using a low-accuracy model in a high-risk situation poses a higher risk than a high-accuracy model in a low-risk situation. </p> <p>Considering all of this, the question to ask here is: what are the risks of employing your AI tool in the way you envision, given its current performance? Ideally, you want it to have as few negative implications as possible on people when it comes to their safety (physical and cyber) and livelihood. If the risks are high, ask if the risk is worth taking. </p> <h3 class="wp-block-heading">DEVELOPMENT RISKS </h3> <p>In some cases, the development of an AI tool itself can cause unwanted trouble, even if not intended. For example, by developing an AI tool that can guess<em> login passwords</em> as an interesting R&D problem can cause undesirable consequences when it lands in the hands of a bad actor. </p> <p>It’s one thing if law enforcement is developing such a “dangerous” tool in a constrained environment to catch predators. It’s another thing altogether if the development team intends to open-source the tool, essentially distributing it to the public. The latter can have many unintended consequences, and developers should be responsible for not just how they will use the tool but also how they<strong> share the tool.</strong></p> <p>The question you want to ask when it comes to development risks is: Have you considered the risks of developing your AI tool using the intended distribution methods?</p> <figure class="wp-block-image"><img decoding="async" src="https://www.opinosis-analytics.com/wp-content/uploads/2023/02/AI-ETHICS-CONSIDERATIONS-1-1024x768.png" alt="" class="wp-image-17371"/></figure> <h2 class="wp-block-heading">Last Word</h2> <p>As AI becomes increasingly integrated into our lives, we must consider its ethical ramifications. In this article, we specifically explored what AI ethics means in creating and using AI-powered tools and the questions to consider for each ethical element. </p> <p>In summary, there are four broad considerations when it comes to AI ethics, and they are:</p> <ul><li><strong>Data Etiquette:</strong> Is the quality of data used to train models known?</li> <li><strong>Explainability:</strong> Can you intuitively explain the AI’s decisions?</li> <li><strong>Usage Etiquette: </strong>Have you considered how AI will fit into your workflow and what the risks are in that scenario?</li> <li><strong>Development Risks</strong>: Have you considered the broader consequences of developing and distributing your AI tool?</li></ul> <p>Each of these considerations focuses on a different angle of how AI can potentially cause harm. In a future article, we’ll explore some of the common ethical challenges of AI systems.</p> <p></p> <hr class="wp-block-separator has-alpha-channel-opacity"/> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p>The post <a href="https://www.opinosis-analytics.com/blog/what-is-ai-ethics/" rel="nofollow">AI Ethics Series: What is AI Ethics</a> appeared first on <a href="https://www.opinosis-analytics.com" rel="nofollow">Opinosis Analytics</a>.</p> <p><a href="https://www.opinosis-analytics.com/blog/what-is-ai-ethics/" target="_blank" rel="noopener">AI Ethics Series: What is AI Ethics</a></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/what-is-ai-ethics/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8600</post-id> </item> <item> <title>GPT-3: What is GPT-3 and what can it do for your business?</title> <link>https://kavita-ganesan.com/what-is-gpt-3/</link> <comments>https://kavita-ganesan.com/what-is-gpt-3/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Tue, 08 Nov 2022 22:51:39 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Foundation]]></category> <category><![CDATA[GPT-3]]></category> <category><![CDATA[nlp]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8377</guid> <description><![CDATA[GPT-3 is a popular AI tool that everyone is raving about. What exactly is GPT-3 and what does it mean for your AI problems? ]]></description> <content:encoded><![CDATA[<p>There’s been a lot of talk about GPT-3 and generative AI in the news, social media, and probably from every AI practitioner or vendor whom you’ve been speaking with lately. </p> <p>Everyone is super excited about the future that such AI tools hold.</p> <p>But what exactly is this AI technology specifically and what does it mean for your business and your AI problems? Let’s explore!</p> <nav class="wp-block-stackable-table-of-contents stk-block-table-of-contents stk-block stk-9b999f4 stk-block-background" data-block-id="9b999f4"><style>.stk-9b999f4 li,.stk-9b999f4 ul li a,.stk-9b999f4 ol li a{font-style:normal !important;font-family:"Poppins",Sans-serif !important}.stk-9b999f4{background-color:var(--ast-global-color-4) !important}.stk-9b999f4:before{background-color:var(--ast-global-color-4) !important}.stk-9b999f4 ol{list-style-type:decimal !important}.stk-9b999f4 .stk-table-of-contents__table{column-count:1 !important}</style><ol class="stk-table-of-contents__table"><li><a href="#what-is-gpt-3">What is GPT-3?</a></li><li><a href="#what-can-gpt-3-do">What Can GPT-3 Do?</a></li><li><a href="#the-business-benefits-of-gpt-3">The Business Benefits of GPT-3 </a></li><li><a href="#is-traditional-ml-going-away-because-of-gpt-3">Is traditional ML going away because of GPT-3?</a></li><li><a href="#what-are-the-risks-of-gpt-3">What are the risks of GPT-3? </a></li><li><a href="#gpt-3-examples">GPT-3 Examples</a></li><li><a href="#gpt-3-key-takeaways">GPT-3 Key Takeaways</a></li><li><a href="#undefined">Keep Learning & Succeed With AI</a></li><li><a href="#further-reading">Resources</a></li></ol></nav> <p></p> <h2 class="wp-block-heading" id="what-is-gpt-3">What is GPT-3?</h2> <p>GPT-3 is a large <strong>language model </strong>developed by Open AI. It’s the successor of Open AI’s older language model, GPT-2 which was much smaller in comparison. </p> <p>So, what’s a language model? A language model is a probability distribution over sequences of words learned from data. This probability distribution can then be used to <em>complete sentences</em>, <em>validate sentence correctness</em>, <em>validate speech recognition predictions</em>, t<em>ranslate one language to another</em>, and much more. </p> <p></p> <p>As you can see, language models are pretty powerful, and in concept, this is not new. Language models have been around for decades.</p> <p></p> <p>Here’s an example of how a language model completes a sentence:</p> <blockquote class="wp-block-quote is-style-plain is-layout-flow wp-block-quote-is-layout-flow"><pre class="wp-block-code"><code class="">The car is about to ______ crash => probability 0.08 stop => probability 0.92 Predicted answer: stop</code></pre></blockquote> <p>Leveraging this general language model concept, GPT-3 is a gigantic language model capable of generating <em>sequences of words</em>, <em>code</em>, <em>translations</em>, <em>summaries</em>, or other types of data, starting from a source input, called the prompt.</p> <p></p> <p>Traditionally, language models have been trained on small datasets as it’s computationally expensive to train large language models. However, GPT-3, is trained on much of the Web, books, and Wikipedia data, which boils down to it being trained on billions of words. Further, GPT-3 is trained using a very deep and sophisticated <a href="https://www.ibm.com/cloud/learn/neural-networks">neural network</a>, helping it learn complex relationships between words.</p> <p></p> <p>This sort of training is not something we can easily replicate as it can cost millions of dollars for every training iteration. In fact, it cost approximately <strong>4.6 million dollars</strong> to train GPT-3 using a Tesla V100 cloud instance over 9 days. But what this level of sophistication means is that GPT-3 can answer all sorts of questions and complete sophisticated tasks with little hand-holding. You can think of GPT-3 as a <strong>super-intelligent Q&A machine.</strong></p> <p></p> <h2 class="wp-block-heading" id="what-can-gpt-3-do">What Can GPT-3 Do?</h2> <p>Some of the capabilities of GPT-3 include:</p> <ul><li>Predicting categories on text data</li> <li>Generating relevant source code based on description alone</li> <li>Extracting pertinent information from unstructured data to make it more structured</li> <li>Becoming your therapy chatbot</li> <li>Translating text in one language into several others that it understands </li> <li>Writing paragraphs of content with a prompt </li> <li>Rewriting article headlines</li> <li>Spelling correction </li> <li>And much much more</li></ul> <p></p> <h2 class="wp-block-heading" id="the-business-benefits-of-gpt-3">The Business Benefits of GPT-3 </h2> <p>So, what is the benefit of GPT-3 for business applications? </p> <p>In short: one model that can complete <strong>multiple tasks</strong>. Years ago, we had to develop a single specialized model for every task that we were looking to solve with AI. We needed the training data, the appropriate ML algorithm, and a data scientist.</p> <p>But with large language models like GPT-3, for many tasks, you can leverage this single model by briefly <em>teaching</em> the model with examples of what types of output to produce. For certain tasks, you don’t even need that. You can just describe the task and provide the input and GPT-3 will generate relevant output. So, almost anyone can perform the AI “development”. </p> <p>For example, if you’re performing a classification task, you can prime the model on the types of expected categories. If you’re wanting generated content, you can tell what type of content you’re expecting. So it essentially democratizes AI development and makes it less time-consuming. </p> <p>Imagine developing a sentiment classifier with just 5 prompts. Is this too good to be true? The only way to know if it holds water on your data is to evaluate, evaluate, and evaluate. You will never run away from evaluation no matter how sophisticated the model as I repeatedly talk about in my <a href="https://kavita-ganesan.com/the-business-case-for-ai/">book</a>.</p> <h2 class="wp-block-heading" id="is-traditional-ml-going-away-because-of-gpt-3">Is traditional ML going away because of GPT-3?</h2> <p>No. Task-specific models, smaller language models and classical ML is not going anywhere anytime soon. GPT-3 only works on tasks that it understands well or tasks that you can make it understand (see examples below). If you have highly domain-specific tasks, you’ll still have to build specialized models that are fine-tuned solely for those tasks. </p> <p>This only means that it’s going to get much easier to develop ML solutions for certain well-understood tasks. Or these models can be used to generate supplementary input for your specialized ML tasks.</p> <h2 class="wp-block-heading" id="what-are-the-risks-of-gpt-3">What are the risks of GPT-3? </h2> <p>Now let’s talk about the hard stuff. While GPT-3 has great potential, we still need to consider its broader implications for your AI applications and business. Some of the risks of GPT-3 include:</p> <ul><li><strong>Bias propagation</strong>—As GPT-3 was predominantly trained on Web data, it has learned both the good and the bad of the Web. This means any embedded bias, errors in data, and non-factual content can easily seep into your applications. </li> <li><strong>Potential plagiarism</strong>—Having knowledge of the entire Web (almost) also means that GPT-3 can spit out content from various sources word-for-word without attribution. So, if you see familiar content within a third-party application, don’t be surprised—it could be YOUR content. Unfortunately, you may not be able to claim plagiarism and there’s not much we can do about it as the model is already open for public usage.</li> <li><strong>Unpredictable performance</strong>—GPT-3 is essentially a language generator that can multitask. And because it’s not fine-tuned for your application-specific task, its performance on a single “narrow” task may be unpredictable. One small glitch can result in erroneous output. </li> <li><strong>Hallucinations</strong>—As GPT-3 computes the probability of generating meaningful output, it could very well stitch together unrelated concepts that IT thinks make sense. This could end up being nonfactual and inaccurate information. If you’re using GPT-3 to generate content, you should validate the facts produced especially on uncommon or time-dependent topics, and topics that are subject to interpretation. </li></ul> <p>These risks are real and people are already raising these issues in various <a href="https://www.reddit.com/r/OpenAI/comments/hypo3o/how_do_i_tell_if_gpt3_is_plagiarizing/">formats</a>.</p> <h2 class="wp-block-heading" id="gpt-3-examples">GPT-3 Examples</h2> <p>Here are two examples of GPT-3 in action. </p> <p></p> <h3 class="wp-block-heading" id="sentence-correction">Sentence Correction</h3> <p>In this example, GPT-3 is asked to edit English sentences.</p> <figure class="wp-block-image size-full"><img decoding="async" width="808" height="554" src="https://kavita-ganesan.com/wp-content/uploads/gpt3-sentence-correction.png" alt="" class="wp-image-8388" srcset="https://kavita-ganesan.com/wp-content/uploads/gpt3-sentence-correction.png 808w, https://kavita-ganesan.com/wp-content/uploads/gpt3-sentence-correction-300x206.png 300w, https://kavita-ganesan.com/wp-content/uploads/gpt3-sentence-correction-150x103.png 150w, https://kavita-ganesan.com/wp-content/uploads/gpt3-sentence-correction-768x527.png 768w" sizes="(max-width: 808px) 100vw, 808px" /></figure> <p class="has-text-align-center">GPT-3: Sentence correction</p> <p></p> <h3 class="wp-block-heading" id="sentiment-classification">Sentiment Classification</h3> <p>In this example, GPT-3 is given examples of how to classify sentences, and then it does it on the last task. </p> <figure class="wp-block-image alignleft size-full"><img decoding="async" width="886" height="290" src="https://kavita-ganesan.com/wp-content/uploads/gpt-3-sentiment.png" alt="" class="wp-image-8391" srcset="https://kavita-ganesan.com/wp-content/uploads/gpt-3-sentiment.png 886w, https://kavita-ganesan.com/wp-content/uploads/gpt-3-sentiment-300x98.png 300w, https://kavita-ganesan.com/wp-content/uploads/gpt-3-sentiment-150x49.png 150w, https://kavita-ganesan.com/wp-content/uploads/gpt-3-sentiment-768x251.png 768w" sizes="(max-width: 886px) 100vw, 886px" /></figure> <figure class="wp-block-image size-full"><img loading="lazy" decoding="async" width="668" height="148" src="https://kavita-ganesan.com/wp-content/uploads/gpt3-sentiment-analysis.png" alt="" class="wp-image-8390" srcset="https://kavita-ganesan.com/wp-content/uploads/gpt3-sentiment-analysis.png 668w, https://kavita-ganesan.com/wp-content/uploads/gpt3-sentiment-analysis-300x66.png 300w, https://kavita-ganesan.com/wp-content/uploads/gpt3-sentiment-analysis-150x33.png 150w" sizes="(max-width: 668px) 100vw, 668px" /></figure> <p class="has-text-align-center">GPT-3: Sentiment orientation prediction</p> <p></p> <p></p> <h2 class="wp-block-heading" id="gpt-3-key-takeaways">GPT-3 Key Takeaways</h2> <ul><li>GPT-3 is a large language model that can help you complete multiple tasks with little to no supervision. </li> <li>GPT-3 cannot solve every AI problem. It’s only as good as the prompts that you feed it and tasks that it understands and may not work well for specialized tasks such as predicting market movements.</li> <li>While GPT3 has great potential, it has its fair share of problems just like any ML model. Some of the potential problems include bias propagation, hallucinations, and unpredictable performance. </li> <li>Just as with any AI solution, evaluation is critical to the success of every initiative and there’s no exception to GPT-3.</li></ul> <p>That’s all for this article!</p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p></p> <h2 class="wp-block-heading" id="further-reading">Resources</h2> <ul><li><a href="GPT-3: Its Nature, Scope, Limits, and Consequences">GPT-3: Its Nature, Scope, Limits, and Consequences</a></li> <li><a href="https://www.toptal.com/gpt-3">Looking for GPT-3 developers, checkout Toptal</a></li></ul>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/what-is-gpt-3/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8377</post-id> </item> <item> <title>What are Precision & Recall in Machine Learning?</title> <link>https://kavita-ganesan.com/precision-and-recall-machine-learning/</link> <comments>https://kavita-ganesan.com/precision-and-recall-machine-learning/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Thu, 27 Oct 2022 01:42:48 +0000</pubDate> <category><![CDATA[AI Foundation]]></category> <category><![CDATA[precision]]></category> <category><![CDATA[recall]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8367</guid> <description><![CDATA[Precision and recall are commonly used terms to assess machine learning model performance. Learn what precision and recall are from a business perspective.]]></description> <content:encoded><![CDATA[<p>Precision and recall are commonly used metrics to measure the performance of machine learning models or AI solutions in general. It helps understand how well models are making predictions.</p> <p>Let’s use an email SPAM prediction example. Say you have a model that looks at an email and decides whether it’s SPAM or NOT SPAM. To see how well it’s doing, you want to compare it with human-generated labels, which we will call the <em>actual labels</em>. </p> <p>To demonstrate this, the table below shows you some actual labels and the machine (model) predicted labels. Now we’ll assume that the <strong><em>spam</em></strong> prediction is positive, and the not spam prediction is negative.</p> <figure class="wp-block-table aligncenter is-style-regular"><table class="has-background has-fixed-layout" style="background:linear-gradient(135deg,rgb(238,238,238) 26%,rgb(158,197,226) 100%)"><thead><tr><th><strong>Email ID</strong></th><th><strong>Actual Label</strong></th><th><strong>Machine Predicted Label</strong></th></tr></thead><tbody><tr><td><strong>Email 1</strong></td><td>Spam (positive)</td><td>Spam (positive & correct)</td></tr><tr><td><strong>Email 2</strong></td><td>Spam (positive)</td><td>Not Spam (negative & incorrect)</td></tr><tr><td><strong>Email 3</strong></td><td>Not Spam (negative)</td><td>Spam (positive & incorrect) </td></tr><tr><td><strong>Email 4</strong></td><td>Spam (positive)</td><td>Not Spam (negative & incorrect) </td></tr></tbody></table><figcaption class="wp-element-caption"><em>Email spam predictions with corresponding true predictions.</em></figcaption></figure> <h2 class="wp-block-heading">What is Precision in ML?</h2> <p>Given this, intuitively, <strong>precision </strong>measures the proportion of correct positive predictions. </p> <figure class="wp-block-image aligncenter size-large"><img decoding="async" src="http://www.opinosis-analytics.com/wp-content/uploads/2021/08/precision-and-recall-machine-learning-1024x309.png" alt="" class="wp-image-13868"/><figcaption class="wp-element-caption">How precision is computed</figcaption></figure> <p>As you can see from the table above, out of the 2 <strong>spam</strong> (positive) machine predictions, only 1 is correct. So the precision is 0.5 or 50%.</p> <h2 class="wp-block-heading">What is Recall in ML?</h2> <p><strong>Recall </strong>measures the proportion of actual positive labels correctly identified by the model. </p> <figure class="wp-block-image aligncenter size-large"><img decoding="async" src="http://www.opinosis-analytics.com/wp-content/uploads/2021/08/recall-computing-in-machine-learning-1024x316.png" alt="recall computing in machine learning" class="wp-image-13870"/><figcaption class="wp-element-caption">How recall is computed</figcaption></figure> <p>From the table above, notice that we have 3 actual labels that are positive, and out of that only one is correctly captured by the model. So the recall is 0.33 or 33%. </p> <p>All in all, in the SPAM prediction example, precision is 50% and recall is 33%. </p> <h2 class="wp-block-heading">What Message Do Precision and Recall Convey?</h2> <p>What precision measures at a high level is <em>correctness</em>. What recall measures at a high level is <em>coverage</em>. For example, if precision is 98% it means that when the model says the prediction is positive, the prediction is likely accurate. A model can be overly conservative and only make limited positive predictions, resulting in high precision. In other words, it fails to make sufficient positive predictions. This is why you also need to consider recall—to ensure you’re capturing sufficient actual positives. </p> <p>When it comes to recall, a high recall means that the model can capture most of the positive predictions. But if a model says everything is positive regardless of underlying reasoning, the recall will be artificially high and close to perfect. That’s why you need to balance between precision and recall. You want accurate predictions, but at the same time not at the cost of missing out on too many positive predictions (false negative predictions). Ideally, you want sufficiently high precision and recall. </p> <h2 class="wp-block-heading">Summary</h2> <p>In summary, <strong>precision </strong>measures the proportion of correct positive predictions, and <strong>recall</strong> measures the coverage of actual positive labels. For a model to be considered “good” both precision and recall must be at acceptable levels. In the end, what’s acceptable depends on the application. </p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p></p> <h2 class="wp-block-heading">Recommended Reading:</h2> <ul><li><a href="https://kavita-ganesan.com/how-to-compute-precision-and-recall-for-a-multi-class-classification-problem/">How to compute precision and recall</a> </li></ul> <div class="inherit-container-width wp-block-group is-layout-constrained wp-block-group-is-layout-constrained"><div class="wp-block-group__inner-container"><div class="wp-block-columns is-layout-flex wp-container-core-columns-is-layout-1 wp-block-columns-is-layout-flex"><div class="wp-block-column is-vertically-aligned-center is-layout-flow wp-block-column-is-layout-flow" style="flex-basis:100%"></div></div></div></div>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/precision-and-recall-machine-learning/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8367</post-id> </item> <item> <title>The Business Case For AI: A Review by Customer Contact Week Conference & Magazine</title> <link>https://kavita-ganesan.com/the-business-case-for-ai-review-customer-contact-week/</link> <comments>https://kavita-ganesan.com/the-business-case-for-ai-review-customer-contact-week/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Wed, 26 Oct 2022 17:21:15 +0000</pubDate> <category><![CDATA[The Business Case For AI]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8319</guid> <description><![CDATA[The Business Case for AI is perfect for AI-fanatics and AI-skeptics alike. From defining AI and its various forms, to explaining exactly how companies should approach AI to see the results they want, Kavita Ganesan’s no-fluff book helps make the AI conversation a whole lot easier.]]></description> <content:encoded><![CDATA[<p>“Stop using AI.”</p> <p>This is how Dr. Kavita Ganesan, an AI expert since 2005, begins her book The Business Case for AI. In a refreshingly direct tone, Ganesan goes on to deliver the news that, yes, you probably need to rethink your use of AI and, no, it does not need to be this difficult or this expensive. </p> <p>While AI is a necessary tool for businesses to remain competitive, many find themselves worried about the investment, and the consequences– what if I’m using AI to solve the wrong problems? Could it really take away my job or my employees’ jobs?</p> <p>Ganesan is a sharp but assuring voice in a field full of questions like these. “These worries are real,” she states before going on to describe exactly what you need to do to assuage them. Throughout the book, Ganesan tells you where you can apply AI, how you can prepare your organization for it, how to invest in the right AI opportunities, and how to measure the success of your AI initiatives, almost making it sound easy. </p> <p>The book does not go into technical language or AI-specific jargon that a CS professional might not be familiar with. Ganesan makes AI, a topic that many feel is unapproachable and a bit overwhelming, something that even<em><strong> the least tech-savvy person can grasp</strong></em>.</p> <p>Ganesan makes <strong>AI sound not only less intimidating but ultimately worthwhile for CS teams in any sector.</strong> She explains that there is an irresistible pull for customers when an experience is personalized, and the most successful companies in the world have utilized the hyper-personalization that only AI can provide. </p> <p>The Business Case for AI is<strong> perfect for AI-fanatics and AI-skeptics alike.</strong> From defining AI and its various forms, to explaining exactly how companies should approach AI to see the results they want, Kavita Ganesan’s no-fluff book helps make the AI conversation a whole lot easier.</p> <blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow"><p>The Business Case for AI is<strong> perfect for AI-fanatics and AI-skeptics alike.</strong> From defining AI and its various forms, to explaining exactly how companies should approach AI to see the results they want, Kavita Ganesan’s no-fluff book helps make the AI conversation a whole lot easier.</p><cite>CCW Conference & Magazine</cite></blockquote> <p></p> <p><strong>Original source: </strong><a href="https://www.customercontactweekdigital.com/ccwomen/whitepapers/four-books-authored-by-women-in-customer-experience">CCW Digital</a></p> <p></p> <p></p> <h2 class="has-text-align-center wp-block-heading"></h2> <div class="wp-block-columns is-layout-flex wp-container-core-columns-is-layout-2 wp-block-columns-is-layout-flex"><div class="wp-block-column is-layout-constrained wp-block-column-is-layout-constrained" style="padding-top:0px;padding-right:0px;padding-bottom:0px;padding-left:0px"><div class="wp-block-stackable-image stk-block-image has-text-align-center stk-block stk-34201b4" data-block-id="34201b4"><style>.stk-34201b4 .stk-img-wrapper img{object-fit:fill !important}.stk-34201b4 .stk-img-wrapper img{border-radius:14px !important}</style><figure class="stk-img-wrapper stk-image--shape-stretch"><img loading="lazy" decoding="async" class="stk-img wp-image-7315" src="https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565.png" width="3260" height="2480" alt="AI strategy book" srcset="https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565.png 3260w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-300x228.png 300w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-1024x779.png 1024w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-150x114.png 150w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-768x584.png 768w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-1536x1168.png 1536w, https://kavita-ganesan.com/wp-content/uploads/040-tx-e1648764756565-2048x1558.png 2048w" sizes="(max-width: 3260px) 100vw, 3260px" /></figure></div></div> <div class="wp-block-column is-layout-constrained wp-block-column-is-layout-constrained" style="flex-basis:60%"><div class="wp-block-stackable-heading stk-block-heading stk-block stk-dbf0db7" id="strong-get-your-copy-today-strong" data-block-id="dbf0db7"><h2 class="stk-block-heading__text"><strong>Get Your Copy Today</strong>!</h2></div> <div class="wp-block-stackable-text stk-block-text stk-block stk-f1e3335" data-block-id="f1e3335"><p class="stk-block-text__text"><em>Whether you want to jumpstart your AI strategy, manage your AI initiatives for better outcomes, or simply find inspiration for your own AI and machine learning applications, The Business Case for AI is your blueprint for AI success.</em></p></div></div></div> <div class="wp-block-buttons is-content-justification-center is-layout-flex wp-container-core-buttons-is-layout-1 wp-block-buttons-is-layout-flex"><div class="wp-block-button"><a class="wp-block-button__link" href="http://amazon.aibusinesscasebook.com">BUY ON AMAZON</a></div></div> <div class="wp-block-stackable-feature alignfull stk-block-feature stk-block stk-27ec623 is-style-horizontal" data-block-id="27ec623"><style>.stk-27ec623-container{display:flex !important}.stk-27ec623-container{align-items:flex-start !important}</style><div class="stk-content-align stk-27ec623-column alignwide stk-container stk-27ec623-container stk--no-background stk--no-padding"><div class="stk-inner-blocks stk-block-content stk-row"><div class="wp-block-stackable-column stk-block-column stk-block-column--v2 stk-column stk-block stk-accbde2" data-block-id="accbde2"><style>.stk-accbde2{align-self:center !important}.stk-accbde2-container{display:flex !important}.stk-accbde2-inner-blocks{justify-content:flex-start !important}.stk-accbde2-container{flex-direction:column !important}</style><div class="stk-column-wrapper stk-block-column__content stk-container stk-accbde2-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-accbde2-inner-blocks"></div></div></div></div></div></div> <p></p> <div class="wp-block-stackable-image-box stk-block-image-box stk-hover-parent stk-block stk-a3551ad is-style-default" data-block-id="a3551ad"><div class="stk-block-content stk-inner-blocks has-text-align-center stk-row stk-block-image-box__content"><div class="wp-block-stackable-column stk-block-column stk-block-column--v2 stk-column stk-block stk-6cafb71" data-block-id="6cafb71"><style>.stk-6cafb71{display:flex !important}.stk-6cafb71{align-items:center !important}</style><div class="stk-column-wrapper stk-block-column__content stk-container stk-6cafb71-container stk--no-background stk--no-padding"><div class="stk-block-content stk-inner-blocks stk-6cafb71-inner-blocks"><div class="wp-block-stackable-subtitle stk-block-subtitle stk-block stk-8888f21" data-block-id="8888f21"><style>.stk-8888f21{margin-bottom:8px !important}.stk-8888f21{opacity:0 !important}:where(.stk-hover-parent:hover,.stk-hover-parent.stk--is-hovered) .stk-8888f21{opacity:1 !important}.stk-8888f21 .stk-block-subtitle__text{color:#FFFFFF !important}</style><p class="stk-block-subtitle__text stk-subtitle has-text-color has-white-color">Subtitle for This Block</p></div></div></div></div></div></div>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/the-business-case-for-ai-review-customer-contact-week/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8319</post-id> </item> <item> <title>What went wrong with Tay, the Twitter bot that turned racist?</title> <link>https://kavita-ganesan.com/tay-twitter-bot/</link> <comments>https://kavita-ganesan.com/tay-twitter-bot/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Fri, 21 Oct 2022 16:09:00 +0000</pubDate> <category><![CDATA[AI Fails]]></category> <category><![CDATA[AI For Business]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8285</guid> <description><![CDATA[Why Tay the Twitter bot from 2016 failed and what we can learn from it. How can we generalize this to ML development best practices?]]></description> <content:encoded><![CDATA[<p>Of late, we’ve been hearing about <strong>Twitter bots</strong> in the news due to the whole saga of <a href="https://techcrunch.com/2022/10/04/elon-intends-buy-twitter/">Elon Musk buying Twitter</a>. One of the reasons the deal took so long to pan out was Musk’s concerns about the number of <strong><em>spam</em></strong> bots running rampant on the platform. While Musk <a href="https://www.wsj.com/podcasts/the-journal/elon-musk-doesnt-want-to-buy-twitter-anymore/72ac2322-bb3e-4d41-8237-16e8e115d3d5">believes that bots make up more than 20%</a> of accounts on Twitter, Twitter states that the number of bots on its platform is marginal.</p> <p>So, what’s this <em>Twitter bot </em>thing?</p> <p>A Twitter bot is essentially a Twitter account controlled by software automation rather than an actual human. It is programmed to behave like regular Twitter accounts, liking Tweets, retweeting, and engaging with other accounts. </p> <p>Twitter bots can be helpful for specific use cases, such as sending out critical alerts and announcements. On the flip side, they can also be used for nefarious purposes, such as starting a disinformation campaign. These bots can also turn nefarious when “programmed” incorrectly. </p> <p>This is what happened with Tay, an AI Twitter bot from 2016. </p> <p>Tay was an experiment at the intersection of ML, NLP, and social networks. She had the capacity to Tweet her “thoughts” and engage with her growing number of followers. While other chatbots in the past, such as <a href="https://web.njit.edu/~ronkowit/eliza.html">Eliza</a>, conducted conversations using narrow scripts, Tay was designed to learn more about language over time from its environment, allowing her to have conversations about any topic. </p> <p>In the beginning, Tay engaged harmlessly with her followers with benign Tweets. However, after a few hours, Tay started tweeting highly <em>offensive</em> things, and as a result, she was shut down just sixteen hours after her launch.</p> <p>You may wonder how can such an “error” happen so publicly. Wasn’t this bot tested? Weren’t the researchers aware that this bot was an evil, racist bot before releasing it? </p> <p>These are valid questions. To get into the crux of what went wrong, let’s study some of the problems in detail and try to learn from them. This will help us all see how to handle similar challenges when deploying AI in our organizations. </p> <h2 class="wp-block-heading">Data</h2> <p>Data is often a big reason why AI models fail. In the case of Tay, shortly after her release, Twitter trolls started engaging the bot with racist, misogynistic, and anti-Semitic language. And because Tay had the capacity to learn as she went, it meant that she internalized some of the language taught by the trolls. Tay just repeated some of this language. Tay uttered bad language because she was fed bad data.</p> <p><strong>Take note: </strong>Poor-quality, prejudiced, or downright bad training data can significantly impact how machine learning models behave. You train ML models with nonrepresentative data, and they will churn out biased predictions. If you starve models of data or feed models incomplete data, they will make random predictions instead of meaningful ones. Questionable learning/training data = questionable output. </p> <blockquote class="wp-block-quote has-text-align-left is-style-plain is-layout-flow wp-block-quote-is-layout-flow" style="font-style:normal;font-weight:500"><p></p> <p>Questionable training data = questionable ML model output </p> <p></p></blockquote> <figure class="wp-block-image aligncenter size-full"><a href="https://arstechnica.com/"><img loading="lazy" decoding="async" width="628" height="308" src="https://kavita-ganesan.com/wp-content/uploads/image-13.png" alt="" class="wp-image-8304" srcset="https://kavita-ganesan.com/wp-content/uploads/image-13.png 628w, https://kavita-ganesan.com/wp-content/uploads/image-13-300x147.png 300w, https://kavita-ganesan.com/wp-content/uploads/image-13-150x74.png 150w" sizes="(max-width: 628px) 100vw, 628px" /></a><figcaption class="wp-element-caption">Example of Tay’s Tweet (source: <a href="https://arstechnica.com/">https://arstechnica.com/</a>)</figcaption></figure> <p></p> <figure class="wp-block-image aligncenter size-full is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/image-14.png" alt="Tay twitter bot" class="wp-image-8306" width="576" height="271" srcset="https://kavita-ganesan.com/wp-content/uploads/image-14.png 748w, https://kavita-ganesan.com/wp-content/uploads/image-14-300x141.png 300w, https://kavita-ganesan.com/wp-content/uploads/image-14-150x71.png 150w" sizes="(max-width: 576px) 100vw, 576px" /><figcaption class="wp-element-caption">Example of Tay’s Tweet (source: <a href="https://twitter.com/geraldmellor/status/712880710328139776?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E712880710328139776%7Ctwgr%5E19f193e907faca963bca95676c139db39fd3cce6%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Farstechnica.com%2Finformation-technology%2F2016%2F03%2Fmicrosoft-terminates-its-tay-ai-chatbot-after-she-turns-into-a-nazi%2F">Twitter.com</a>)</figcaption></figure> <h2 class="wp-block-heading">Design </h2> <p>While we don’t often relate model or solution design to erratic model behaviors, it’s often more common than you think. By design, Tay continuously learned from external input (i.e., the environment). Among all the benign Tweets that Tay consumed from her environment were also <em>abrasive</em> Tweets. The more abrasive Tweets Tay saw, the more she learned that those were typical types of responses to Tweet. </p> <p>This is true of any ML model. The dominant patterns influence the predictions of the ML models. Fortunately, it’s not necessary for ML models to learn continuously from their environment. ML models can learn from controlled data. So, Tay’s design itself was risky. </p> <p><strong>Take note:</strong> The design of your ML models impacts how it behaves in reality. So, when designing ML systems, developers and business stakeholders should consider the different ways in which the system can fail, operate suboptimally, be breached, and adjust the design accordingly. In the end, you need a fail-safe plan. </p> <p>In the case of Tay, such thinking early on would’ve made clear that not all Tweet engagements would be benign. There could be bad actors tweeting and engaging in a highly offensive manner, not far-fetched at all from reality. The realization that the bot could be consuming bad data may have stopped the team from using data from other Twitter accounts. They may also have considered consuming data from approved Twitter accounts. </p> <blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow"><p>The design of your ML models impacts how it behaves in reality. </p></blockquote> <h2 class="wp-block-heading">Testing</h2> <p>One of the key steps in the machine learning development lifecycle is <em>testing</em>—not just during development, but testing right before full deployment. I call this <em>post-development testing</em> (PDT). </p> <figure class="wp-block-image size-large"><img loading="lazy" decoding="async" width="1024" height="606" src="https://kavita-ganesan.com/wp-content/uploads/post-development-testing-1024x606.png" alt="post-development-testing" class="wp-image-5831" srcset="https://kavita-ganesan.com/wp-content/uploads/post-development-testing-1024x606.png 1024w, https://kavita-ganesan.com/wp-content/uploads/post-development-testing-300x178.png 300w, https://kavita-ganesan.com/wp-content/uploads/post-development-testing-150x89.png 150w, https://kavita-ganesan.com/wp-content/uploads/post-development-testing-768x455.png 768w, https://kavita-ganesan.com/wp-content/uploads/post-development-testing-1536x910.png 1536w, https://kavita-ganesan.com/wp-content/uploads/post-development-testing.png 1756w" sizes="(max-width: 1024px) 100vw, 1024px" /></figure> <p class="has-text-align-center"><strong>The ML Development Life Cycle</strong></p> <p></p> <p>In the case of Tay, It’s unclear how much PDT went on before releasing the bot, but obviously, it wasn’t enough! Had Tay been subjected to different types of tweet engagements during PDT, the dangers of releasing Tay would’ve become obvious.</p> <p><strong>Take note: </strong>In practice, PDT is often overlooked due to a rush to release a new feature or product. It’s often assumed that if a model works well during development, it will naturally perform well in practice. Sadly, that’s not always the case. So, take note that PDT is critical when it comes to AI deployment.</p> <p></p> <p>During PDT, you can stress test your AI solution to find points of failure. In the case of Tay, subjecting it to different types of Twitter users (e.g., trolls, benign users, and passive aggressives) could’ve surfaced risky behaviors of the bot. PDT can also help evaluate your solution’s impact on relevant business metrics. For example, suppose your business metric measures speed improvement in completing a particular task. PDT can give you early insights into such metrics. </p> <blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow"><p>During PDT, you can stress test your AI solution to find points of failure. PDT can also help evaluate your solution’s impact on relevant business metrics. </p></blockquote> <h2 class="wp-block-heading">Monitoring</h2> <p>Another critical component in the ML development lifecycle is monitoring after deployment. With Tay, monitoring the bot’s behavior eventually led to it being shut down within 24 hours of its release (side note: negative press also had a hand in it). If the bot hadn’t been monitored long after its release, this could’ve led to a whole lot more negative press and many more groups being offended. </p> <p><strong>Take note: </strong>While model monitoring is often done as an afterthought, it should be prioritized before its release to end users. The initial weeks after a model’s release is the most crucial, as unpredictable behaviors not seen during testing could emerge. </p> <blockquote class="wp-block-quote is-layout-flow wp-block-quote-is-layout-flow"><p>The initial weeks after a model’s release is the most crucial, as unpredictable behaviors not seen during testing could emerge. </p></blockquote> <p></p> <h2 class="wp-block-heading">Summary</h2> <p>While what went wrong with Tay may be surprising and intriguing to many, from a machine learning best practices perspective, Tay’s behavior could’ve been predicted. Tay’s environment wasn’t always positive, and she was designed to learn from that environment which led to a perfect recipe for a dangerous experiment. </p> <p>So decisions around data, model design, testing, and monitoring are critical to every AI initiative. And this is not just the responsibility of the developers but also the business stakeholders. The more thought we put into each element, the fewer the surprises and the higher the chances of a successful initiative. </p> <p></p> <p>That’s all for now!</p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/tay-twitter-bot/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8285</post-id> </item> <item> <title>3 Strategic Mistakes Leaders Can Easily Avoid When Thinking About AI Integration</title> <link>https://kavita-ganesan.com/ai-leadership-mistakes/</link> <comments>https://kavita-ganesan.com/ai-leadership-mistakes/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Fri, 09 Sep 2022 17:17:45 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Strategy]]></category> <category><![CDATA[ai for leaders]]></category> <category><![CDATA[ai in business]]></category> <category><![CDATA[ai mistakes]]></category> <category><![CDATA[ai strategy]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8204</guid> <description><![CDATA[Rushing through AI integration can result in unintended failures. Learn 3 AI integration mistakes leaders can easily avoid to limit confusion, reduce waste, and truly reap the benefits from AI. ]]></description> <content:encoded><![CDATA[<p>Several years ago, a product manager at a tech company had a <strong>data collection problem</strong>: to scrape software security vulnerability data from multiple web sources, consolidate the vulnerabilities and store them in a database.<br><br>As this was an automation problem relating to data, the product manager (PM) immediately concluded that this was a <strong><em><a href="https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/">machine learning problem</a></em></strong>. The PM then “hired” the company’s data science team to build ML models to solve the problem. </p> <p>The data science team agreed to the <a href="https://kavita-ganesan.com/machine-learning-training-data-generation/">data collection task</a> without making any promises on “models.” They realized that their attempts to educate the PM that this <strong><em>was a simple script</em></strong> (not a sophisticated ML model) would be a losing battle as there was a big internal push to use AI, and the PM was sold on the idea.</p> <p>Several weeks passed, and when the time came to “deploy” the models, there wasn’t a model to be deployed. Just a software script that would continually read specific webpages, heuristically scrape security vulnerability entries and populate them into a database. </p> <p>Although the PM was eventually informed that no ML models were used or necessary, the scraping software was sold to the entire company as a machine learning powered security solution. </p> <p>This is not uncommon. </p> <p>Such <strong>confusion </strong>around AI and where it’s best employed happens more often than we think. In the case of this tech firm, the confusion didn’t do much damage as it was a small project, and the only thing wasted was the data science team’s precious time for that few weeks. </p> <p>In many other situations, the damage from such miscategorization, poor understanding of AI, and the use of wrong resources can be <strong>extremely costly</strong>.</p> <p>Imagine if the data collection problem above was <strong>forced to use machine learning</strong>, although unnecessary. Maintaining an ML solution costs much more than a simple software script. Plus, if the project had gone on for an entire year, the data scientists would have been paid to solve a problem that a single contracted software engineer could’ve solved. More importantly, these data scientists could’ve been working on high-impact AI initiatives. </p> <p><br>Based on this story, let’s narrow down three strategic mistakes leaders can easily avoid to prevent confusion, reduce waste, and ensure that you’re genuinely reaping the<em> benefits from AI.</em></p> <p></p> <h2 class="wp-block-heading">3 Mistakes Leaders Can Avoid When Thinking About AI Integration</h2> <figure class="wp-block-image aligncenter size-large"><img loading="lazy" decoding="async" width="1024" height="753" src="https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-1024x753.png" alt="ai leadership mistakes to avoid" class="wp-image-8240" srcset="https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-1024x753.png 1024w, https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-300x221.png 300w, https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-150x110.png 150w, https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-768x565.png 768w, https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid-1536x1130.png 1536w, https://kavita-ganesan.com/wp-content/uploads/ai-leadership-mistakes-to-avoid.png 1794w" sizes="(max-width: 1024px) 100vw, 1024px" /></figure> <p class="has-text-align-center">Summary of AI leadership mistakes to avoid</p> <h3 class="wp-block-heading"><strong>#1: Expecting “Others” to Understand AI</strong></h3> <p>In 2018, industry research firm Gartner made a bold prediction—that 85% of AI projects will “not deliver.” This is a shocking prediction, given how important AI has become in recent years. </p> <p>One reason for this prediction is confusion among leaders on what AI is and what it can do.</p> <p>It’s a given that your technical teams need to understand AI. However, executives, technology leaders, and product managers looking to make AI an integral part of their business should also be well-versed with the technology. </p> <p>We’re not talking about getting into AI model development. Still, you need to know AI at the <em>right level </em>to be comfortable exploring the possibility of using AI to solve business problems. </p> <p>Further, this AI knowledge can be handy in several ways.</p> <ul><li><strong>Closing AI adoption gaps: </strong>Once you understand AI, you’ll start seeing the building blocks for preparing your organization for its adoption. You’ll start noticing gaps in your company infrastructure, cultural readiness, and talent pool, allowing you to develop strategies to lay the necessary foundation.</li><li><strong>Vendor selection and hiring: </strong>The AI knowledge will also help you when talking to AI vendors and job candidates, where you’ll be able to ask the right questions, separate the good from the bad, and make sound purchase and hiring decisions.</li><li><strong>Maximize investments:</strong> A breadth of AI understanding will also help you use the proper thought process and frameworks in evaluating which problems would benefit the most from AI, helping you solve the rest of the problems with alternative approaches. With this, you’re increasing the odds of seeing meaningful results from AI.</li></ul> <p></p> <h5 class="has-ast-global-color-4-background-color has-background wp-block-heading"><strong><span style="color: #ad2033;" class="stk-highlight">Take Action:</span></strong> If you’re a leader new to AI, <strong>start by building a foundation</strong> around understanding AI use cases, what it is, and what makes AI initiatives different from traditional software engineering. Understanding the <strong>misconceptions of the field</strong> and how to <strong>spot opportunities</strong> will also significantly help identify high-impact use cases. <br><br>You can get some of this information by reading <a href="https://www.google.com/search?q=ai+business+books">relevant books</a> as well as industry reports from big consulting firms. Attending AI leadership seminars and presentations can also be helpful. Podcasts? I wouldn’t recommend podcasts to build your foundation. The scattered nature of podcasts can be confusing and should be supplemental knowledge once you have a general foundation.</h5> <h2 class="wp-block-heading"> </h2> <h3 class="wp-block-heading"><strong>#2: Expecting a Quick Financial Return from AI </strong></h3> <p>Yes, AI has the promise of cost savings and boosting revenues. Although you may observe an immediate financial impact for some problems, in most cases, you may never see a noticeable financial impact from AI, just from a single initiative. </p> <p>It may take multiple related initiatives coming together to change your financial trajectory, or it’s something you’ll observe over the long haul. </p> <p>So, when it comes to the ROI of AI, you need to focus on the benefits of employing AI (in the short term and long term). Ask these questions:</p> <ul><li>What <strong>immediate pain point</strong> would the AI solution ease for your organization?</li><li>What <strong>benefits </strong>would you see by addressing the pain point?</li><li>What’s the<strong> added advantage</strong> of an AI solution over a simpler one, such as a manual approach?</li></ul> <p>Answering such questions will clarify why AI is necessary and guide you in tracking the right business metrics</p> <p></p> <h5 class="has-ast-global-color-4-background-color has-background wp-block-heading"><span style="color: #ad2033;" class="stk-highlight"><strong>Take Action:</strong> </span>When you’re looking to track the success of AI, <strong>always start with metrics that tie into its direct impact </strong>first. Once this is well underway and delivering results, track metrics that relate to the longer-term implications, which can take months or even years to observe. </h5> <p></p> <h3 class="wp-block-heading"><strong>#3: Leaving AI Completely in the Hands of Data Scientists</strong></h3> <p>In the rush to adopt AI, companies often start by hiring a team of data scientists. This happens long before leaders understand AI or have an AI strategy.</p> <p>These data scientists are then let loose on the data to discover potential AI opportunities. While several identified projects may be meaningful, many are better suited for publishing a research paper—not so much for creating value for the business.<br></p> <p>This is not entirely the fault of the data scientists. Data scientists newly brought in to solve AI problems for the company will have a limited view of the company’s business challenges. </p> <p>Exploring the data tells them nothing about the process and workflow inefficiencies in the company. Further, your company may not be collecting data for problems that would most benefit from AI. Instead of twiddling thumbs, these data scientists are left with no choice but to tackle “made-up” problems with relevant data.<br></p> <p>On the contrary, business unit leaders, executives, and domain experts deal with the organization’s daily challenges—whether it’s customer complaints, a media coverage issue, or friction in your business processes. </p> <p>These employees should be equally capable of spotting opportunities for automation and workflow augmentation with AI. They should feel empowered to bring relevant business problems for data scientists to solve. </p> <p>For companies to succeed with AI, there must be a deep collaboration between business leaders, domain experts, and their technical counterparts.</p> <p></p> <h5 class="has-ast-global-color-4-background-color has-background wp-block-heading"><strong><span style="color: #ad2033;" class="stk-highlight">Take Action:</span></strong> When you witness <strong>process inefficiencies, repetitive manual tasks, and lagging accuracy</strong> of existing software systems, start taking note. Have your team track current baseline performance numbers and determine if the problem relates to solving a complex decision-making task. Such problems are often great candidates for AI. Involve your technical experts to help study the problem further and determine if AI is a good fit and, if not, recommend alternate approaches. </h5> <p></p> <h2 class="wp-block-heading">How Will You Accelerate Your AI Adoption?</h2> <p>While many leaders believe that the success of AI adoption is in the excellence of their technical teams, in reality, it starts at the top. </p> <p>Executives and functional leaders deal with the everyday challenges that the organization faces. With a good AI understanding, they’re better positioned to <strong>recognize problems </strong>that AI will solve and consequently <strong>fund impactful initiatives.</strong> This, coupled with the <strong>right expectations and success metrics</strong>, will translate to better outcomes for the organization.</p> <p></p> <p></p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-2-color has-ast-global-color-4-background-color has-text-color has-background has-medium-font-size is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:solid"><div class="wp-block-group__inner-container"><div class="inherit-container-width wp-block-group has-ast-global-color-5-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading" style="font-style:normal;font-weight:500">Recommended Reading</h2> <ul class="has-medium-font-size"><li><a href="https://kavita-ganesan.com/how-to-improve-ai-outcomes-in-business/">Even if AI is your goal, why start without it to improve AI success</a></li><li><a href="https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/">15 Common AI Problem Types</a></li></ul></div></div></div></div>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/ai-leadership-mistakes/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8204</post-id> </item> <item> <title>How Sentiment Analysis Keeps Your Brand in Check (and How to Get Started)</title> <link>https://kavita-ganesan.com/sentiment-analysis-business-applications/</link> <comments>https://kavita-ganesan.com/sentiment-analysis-business-applications/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Thu, 25 Aug 2022 16:16:47 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Use Cases]]></category> <category><![CDATA[NLP Applications]]></category> <category><![CDATA[nlp]]></category> <category><![CDATA[sentiment analysis]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8104</guid> <description><![CDATA[We’ve all heard of sentiment analysis, but what exactly is it and what can it do for your brand, your business, and how can you get started with it?]]></description> <content:encoded><![CDATA[<p><em>We’ve all heard of sentiment analysis, but what exactly is it and what can it do for your brand, your business, and how can you get started with it?</em></p> <p></p> <div class="wp-block-uagb-table-of-contents uagb-toc__align-left uagb-toc__columns-1 uagb-block-df53d82c " data-scroll= "1" data-offset= "30" style="" > <div class="uagb-toc__wrap"> <div class="uagb-toc__title"> Table Of Contents </div> <div class="uagb-toc__list-wrap"> <ol class="uagb-toc__list"><li class="uagb-toc__list"><a href="#what-is-sentiment-analysis" class="uagb-toc-link__trigger">What is Sentiment Analysis?</a><li class="uagb-toc__list"><a href="#why-is-sentiment-analysis-important-in-business" class="uagb-toc-link__trigger">Why is Sentiment Analysis Important in Business?</a><li class="uagb-toc__list"><a href="#how-are-businesses-using-sentiment-analysis-real-world-examples" class="uagb-toc-link__trigger">How are Businesses Using Sentiment Analysis? (Real-World Examples)</a><li class="uagb-toc__list"><a href="#how-to-get-started-with-sentiment-analysis" class="uagb-toc-link__trigger">How to Get Started with Sentiment Analysis</a><li class="uagb-toc__list"><a href="#keep-learning-from-me" class="uagb-toc-link__trigger">Keep Learning From Me:</a><li class="uagb-toc__list"><a href="#recommeded-reading" class="uagb-toc-link__trigger">Recommeded Reading</a></ol> </div> </div> </div> <p></p> <h2 class="wp-block-heading">What is Sentiment Analysis?</h2> <p>Sentiment analysis relates to analyzing content such as social media comments, customer feedback, employee feedback, and even facial expressions in images to render sentiment orientation. These sentiments can be as <em>broad </em>as just saying that the specific content is from a <strong>“detractor”</strong> or <strong>“promoter,”</strong> or it can be as detailed as listing out all the <strong>emotions</strong> within the content.</p> <p></p> <p class="has-text-align-center"><img loading="lazy" decoding="async" src="https://lh3.googleusercontent.com/ZJTsL89jkMZcNcqFzqlFEzTAVS4xymQVo-Fca_N6X5NZERsMJW3MRcwKeFtQAuyxvQFyL33Wd_Fh_1jqPsa7bcw5d2dXeu1HYBH64NzpkeU21BGZp6NahgjbryDcSKUXt4yxJZH0ntTtX_k8yqerOcs" width="624" height="371"></p> <p class="has-text-align-center">Predicting fine-grained sentiment in images</p> <figure class="wp-block-image aligncenter"><img decoding="async" src="https://lh4.googleusercontent.com/y1MhZlKK3AwBAHNeP5MX6w07aETHFoYaz-eoagqm6AX8pOaKuHFpe3GFFi8lQyJpfcHeTj2L1_X1AHmVsjImCGtzUQ8Ysyh460eqzAMDleT-wnaf0-dnPdT4JEiQ_UQzSgKGBgwkjYC4MeNs8Q3SR5E" alt="sentiment analysis business application examples"/><figcaption>Predicting sentiment in text data</figcaption></figure> <p class="has-text-align-center"></p> <h2 class="wp-block-heading">Why is Sentiment Analysis Important in Business?</h2> <p>While, on the surface, sentiment analysis can seem like a fancy class project, in actuality, it has many uses in business. Let’s look at some sentiment analysis examples applied to business problems.</p> <ol><li>You can <strong>aggregate customer sentiments from free-form feedback data </strong>and determine if your customers are primarily promoters or detractors. You can then take corrective measures to gradually <em>rebuild trust with the detractors</em> and turn them into promoters.</li><li>You can <strong>keep your online platform clean and free from bullies </strong>by detecting hateful and inappropriate comments.</li><li>You can determine <strong>which employees are demotivated </strong>or about to quit based on their outlook from recent feedback, peer reviews, and manager feedback and provide a constructive path ahead for employees to succeed at the company.</li></ol> <p>Overall, as you can see from these sentiment analysis examples, sentiment analysis is a versatile tool that can help you better understand employees and customers, keep platforms safe, provide customers with a better shopping and product selection experience, and learn from competitor brands. </p> <p>More importantly, when you combine <em>sentiment analysis</em> with other AI-driven technologies such as <em>text summarization,</em> you can get deeper, more powerful insights. </p> <h2 class="wp-block-heading">How are Businesses Using Sentiment Analysis? (Real-World Examples)</h2> <p>Now that we know what sentiment analysis can help accomplish, let’s see how three companies are using sentiment analysis for a specific business purpose. </p> <h3 class="wp-block-heading">GAIL</h3> <p>Great Wolf Lodge (GWL), a chain of resorts and indoor water parks, has expanded its broad digital strategy by using <a href="https://www.cio.com/article/3435118/introducing-gail-great-wolf-lodges-ai-for-pinpointing-guest-sentiment.html">AI to classify customer comments based on sentiment</a>. They developed what they call the <em>Great Wolf Lodge’s Artificial Intelligence Lexicographer (GAIL)</em>.</p> <p>GWL capitalizes on the concept of net promoter score (NPS) to gauge the experience of individual customers. </p> <p></p> <p>Instead of using an NPS score to determine customer satisfaction, GAIL determines if customers are <em>net promoter</em>s, <em>detractor</em>s, or <em>neutral parties </em>based on the free-text responses posted in monthly customer surveys. This is analogous to predicting if the customer sentiment is <em>positive</em>, <em>negative</em>, or <em>neutral</em>. GAIL essentially “reads” the comments and generates an opinion.</p> <figure class="wp-block-image aligncenter size-full"><img loading="lazy" decoding="async" width="850" height="447" src="https://kavita-ganesan.com/wp-content/uploads/image-12.png" alt="sentiment analysis business application examples" class="wp-image-8110" srcset="https://kavita-ganesan.com/wp-content/uploads/image-12.png 850w, https://kavita-ganesan.com/wp-content/uploads/image-12-300x158.png 300w, https://kavita-ganesan.com/wp-content/uploads/image-12-150x79.png 150w, https://kavita-ganesan.com/wp-content/uploads/image-12-768x404.png 768w" sizes="(max-width: 850px) 100vw, 850px" /><figcaption>Detractors, promoters, and how the NPS score is computed</figcaption></figure> <p>Through this effort, the company hopes to understand its guests better and improve the customer experience. For example, by analyzing comments by detractors, Great Wolf Lodge would know areas in their service that need improvement. </p> <p>Analyzing this unstructured data manually would take far too long for humans. However, GAIL can <em>parse this data in seconds </em>and determine whether the author is a net promoter, detractor, or neutral party.</p> <h3 class="wp-block-heading">Meta</h3> <p>Meta—with nearly 1.7 billion daily active users—naturally has content posted on the platform that violates its rules. Among this negative content is <a href="https://www.wired.com/story/facebook-ai-hate-speech-improves-unclear/"><strong>hate speech</strong></a><strong>. </strong>Defining and detecting hate speech is one of the biggest political and technical challenges for Meta and similar platforms. Detecting hate speech is a type of sentiment analysis problem focused on content with overall negative implications. </p> <p>Humans review the AI-flagged posts in the same way as posts reported by users. In fact, the platform removed 9.6 million pieces of content flagged as hate speech in the first quarter of 2020 alone. While the sentiment models alone may not be sufficient to control hate speech on the platform, the tool does capture a massive number of spam posts, significantly reducing the amount of manual work by humans.</p> <p class="has-text-align-center"><img loading="lazy" decoding="async" width="624" height="475" src="https://lh6.googleusercontent.com/IOgGQI_H1HU69ZQOM6cIUxe0w4HGkla3hremcoHR-gt9xwog8TmcE0LdpCZwlGRyidpmJcnPWQ4B9FsY8CAm9TTvIb7ic0wqOtzVDob_VOEE-nLrPKIAKZGxZ13ZQBYd9US4J5eJzpzYZbPrDyk0CEo" alt="text categorization example"></p> <p class="has-text-align-center">The volume of AI-based hate speech removal on Facebook. Source: <a href="https://www.wired.com/story/facebook-ai-hate-speech-improves-unclear/">Wired</a></p> <p>Detecting which content contains hate speech is a complicated problem. AI algorithms must understand the subtle meanings in text and nuances in expressions, analyze the cultural context, and then determine whether it’s offensive without incorrectly penalizing harmless content.</p> <p class="has-text-align-center"><img loading="lazy" decoding="async" alt="text classification example" src="https://lh3.googleusercontent.com/zcnfSCER59kFxzJVZUGRiTaEJW2R8cHOzuTNvYnJFqKQ7vXiWAt1gRSIuwy94i_xX54C3ZFHb0iKWRKczGcebMlsyqqf6c_CEQYbFVdC5zdthtCYAO2U4tZnGY6wzgIqw65lbN-4iN7a7giCTqhK-Ds" width="411" height="193"><br>Example hate speech. Source: <a href="https://arxiv.org/pdf/1804.04257.pdf">arxiv.org</a></p> <h3 class="wp-block-heading">Ocean Spray</h3> <p>When the morning juice market weakened, Ocean Spray, an agricultural cooperative of cranberry and grapefruit growers, sought a new strategy to improve sales. Ocean Spray first needed to understand consumer sentiment and behaviors around cranberry juice better so that they could innovate.</p> <p>Typically such innovation is done with the help of small focus groups of 10-15 people. However, Ocean Spray decided to leverage AI-driven analysis of thousands of online conversations, such as user reviews and tweets around cranberry juice, to really<em> listen at scale. </em></p> <p>Plus, instead of just classifying content like what Meta does, Ocean Spray leveraged <em>themes</em> and <em>opinion summaries</em> to understand consumer sentiment around specific topics. Through this analysis, Ocean Spray understood how consumers were using cranberry juice in real life, giving them ideas on how best to innovate and fill gaps in the marketplace. </p> <p>The research surfaced unexpected customer behaviors. For example, they found that women enjoyed cranberry juice as a substitute drink without the alcohol in place of cocktails. Such insights helped them launch<a href="https://news.oceanspray.com/2017-03-2017-Sip-Back-and-Unwind-With-a-Mocktail-Moment"> two new beverage lines</a>, boosting revenues and helping them get out of an over-saturated segment of the market.</p> <figure class="wp-block-image aligncenter size-large is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-1024x1017.png" alt="" class="wp-image-8112" width="773" height="768" srcset="https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-1024x1017.png 1024w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-300x298.png 300w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-150x150.png 150w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-768x763.png 768w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis.png 1224w" sizes="(max-width: 773px) 100vw, 773px" /><figcaption>A new beverage line by Ocean Spray in direct response to understanding consumer behaviors around cranberry juice. Source: <a href="https://news.oceanspray.com/2017-03-2017-Sip-Back-and-Unwind-With-a-Mocktail-Moment">oceanspray.com</a></figcaption></figure> <p></p> <h2 class="wp-block-heading">How to Get Started with Sentiment Analysis</h2> <p>As you’ve seen in this article, sentiment analysis has many nuances—you can detect sentiments in a sentence, paragraphs of text, and even from facial expressions in images. Further, you have various ways to leverage sentiment information—from using it for new product innovation to improving the customer experience. </p> <p></p> <p>To get started with sentiment analysis, you first need to understand your business <strong>application</strong>. Consider these questions:</p> <ul><li>What would you like to know about your brand, customers, or employees? </li><li>How granular should the information be? </li><li>Do you just need sentiment information, or textual themes and summaries?</li><li><span style="background-color: var(--ast-global-color-5); font-size: 1rem;">Are you planning to integrate the solution into your dashboards or perform an independent analysis?</span></li></ul> <p></p> <p>Let’s take an example. Say you need to understand the general sentiment in your company’s <em>support conversations</em>. You want to learn the ongoing “tone” and “mood” of your customers. Further, you want to visualize this within your dashboards. In such a case, you’d need to employ an <em><strong>emotion classifier</strong></em> to generate predictions on <em>relevant conversations</em>. You can then leverage those sentiments in your dashboards for downstream analysis. </p> <figure class="wp-block-image aligncenter size-full"><a href="https://arxiv.org/pdf/1812.01207v1.pdf"><img loading="lazy" decoding="async" width="698" height="612" src="https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-emotion-classification.png" alt="" class="wp-image-8121" srcset="https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-emotion-classification.png 698w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-emotion-classification-300x263.png 300w, https://kavita-ganesan.com/wp-content/uploads/sentiment-analysis-emotion-classification-150x132.png 150w" sizes="(max-width: 698px) 100vw, 698px" /></a><figcaption>Plutchik’s Wheel of Emotions. The boxed emotions are commonly used for developing emotion classifiers. Source: arxiv.org</figcaption></figure> <p></p> <p>Depending on your sentiment analysis problem, in some cases, you’d have to custom build the classifiers. But for others, you can leverage off-the-shelf tools such as <a href="https://medium.com/google-cloud/sentiment-analysis-using-google-cloud-machine-learning-552be9b9c39b">Google’s Natural Language API</a> or the <a href="https://perspectiveapi.com/">Perspective API</a>.</p> <p></p> <p>Often, for a multi-faceted analysis, you’d have to combine <strong>off-the-shelf tools with custom pipelines and analysis </strong>to help you answer all questions for optimal decision-making. This is what one of my clients did. They combined insights from an independent off-the-shelf text analytics tool such as <strong>Netbase</strong> (extremely pricey, by the way) with custom-built pipelines for a complete market research analysis.</p> <p></p> <p>There are endless possibilities in how you can employ these sentiment analysis tools. But remember to let the application guide the solutions that you’ll employ.</p> <p></p> <p>Now, over to you. What sentiment analysis applications come to mind after reading this article? What tools will you use for your analysis? </p> <p></p> <h2 class="wp-block-heading">Keep Learning From Me:</h2> <ul><li><strong>Join my <a href="/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li><li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li></ul> <p></p> <h2 class="wp-block-heading">Recommeded Reading</h2> <ul><li><a href="https://kavita-ganesan.com/natural-language-processing-applications/">7 NLP Applications in Business</a></li><li><a href="https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/">List of Common AI Problem Types in Business</a></li></ul> <p></p> <p></p> <p> </p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/sentiment-analysis-business-applications/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8104</post-id> </item> <item> <title>Even if AI is Your Goal, Why Starting Without AI Improves Outcomes</title> <link>https://kavita-ganesan.com/how-to-improve-ai-outcomes-in-business/</link> <comments>https://kavita-ganesan.com/how-to-improve-ai-outcomes-in-business/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Fri, 12 Aug 2022 01:24:00 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Strategy]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=8000</guid> <description><![CDATA[In this article, you’ll learn why it’s sometimes wiser to start without AI, even if your goal is to integrate AI into products, services, and workflows. You'll also learn tips for handlings AI non-readiness issues. ]]></description> <content:encoded><![CDATA[<p>When it comes to software automation, many teams turn to AI as their potential answer.</p> <p>AI in the form of machine learning or NLP may be an excellent solution to a problem. But did you know that the best way to start AI initiatives is to start with no AI at all?</p> <p>This may seem counterintuitive, but there’s a simple reason for it.</p> <p>It’s because<em> you may not be ready for AI as a solution</em>. There could be several missing elements that’ll prevent you from seeing success with AI if pursued prematurely.</p> <p>When it comes to AI, it’s not far-fetched to say that several critical stars need to align to get results from initiatives from a usability perspective.</p> <p>Let’s look at the reasons why it may be wiser to hold off on AI than to start with it and hit roadblocks and, later, tips for how you can eliminate those roadblocks.</p> <h2 class="wp-block-heading">Why it’s Wiser to Hold Off on AI</h2> <h3 class="wp-block-heading">1: You don’t yet understand the problem you’re solving</h3> <p>Typically, when you’re trying to solve an existing problem with AI, the input, and the desired output are often well understood. You may be looking to AI to improve the accuracy of the existing solution or the speed of completing tasks.</p> <p>But from my experience, a large number of problems that engineering teams and entrepreneurs are solving today are <em>new problems.</em> The problems are weakly defined, and you may not fully understand what your expected output is, let alone what you’re input into the system would be.</p> <p>Take the problem of sentiment prediction. Do you know if you’re looking to predict broad overall sentiments (e.g., positive or negative) or more granular ones like 10% anger and 90% sadness on some given text? Are you looking to feed in paragraphs of text or just a single sentence or short snippets?</p> <p>Yes, this is a design problem. And the right design comes from a good understanding of the problem. Without it, you’ll be battling many design dilemmas. Such design issues, along with the complexity of developing the AI systems, may require you to constantly revamp models to address design changes, introducing confusion and reducing your chances of success with AI.</p> <h3 class="wp-block-heading">2: You may not have the necessary data</h3> <p>As I repeatedly talk about in my <a href="https://kavita-ganesan.com/the-business-case-for-ai/#.YuQTgnbMJPY" target="_blank" rel="noreferrer noopener">book</a>, AI systems demand data. It’s not just data for training models but also data to better understand the problem you’re solving and the expected output from the system.</p> <p>Often, when you have a brand new problem, this data is non-existent. Even for old problems that are being manually solved, the data may not exist or may be available in a non-accessible format. This is exactly what happened with a healthcare client. They were performing a billing annotation task for over eight years, but when came time to automate the process, the data just wasn’t there.</p> <p>Without the right type of data, you won’t really know what problem you’re solving, let alone train a model.</p> <h3 class="wp-block-heading">3: Your users may be skeptical of automation</h3> <p>Let’s face it. People are suspicious of AI, especially those who don’t know what it is and the current state of its capabilities. The moment you talk about automation within employee workflows, some will get uncomfortable and anxious and start to worry about being replaced by the “AI race.”</p> <p>People are also used to their preferred way of doing things. They worry about how their current “efficient” workflow will be affected by the integration of AI. Some think this new AI thing is just a gimmick. This is the exact problem I faced with the healthcare client I mentioned earlier. While the CEO was very enthusiastic about integrating AI in one specific workflow in their business, the employees were not enthusiastic and made that clear.</p> <p>The problem with resistance to using AI is that people may not perceive the solution to be a long-term one. Further, subject matter experts who are suspicious of the automation idea may not be willing to help co-develop a working AI solution, as was the case with my healthcare client. Additional education, training, and buy-in were required to get them to see why automation would make their lives easier. Without customer buy-in, no matter how impressive the AI solution, its existence will be short-lived.</p> <h4 class="wp-block-heading">So, what to give?</h4> <p>Forcing an AI solution on people will not work…in the long term.</p> <p>Starting an AI initiative without data will ensure you hit a dead end.</p> <p>An ill-defined problem will require that you redesign your AI tool over and over again, and this can be expensive.</p> <p>What can you do?</p> <h2 class="wp-block-heading">3 Tips for Handling AI Non-Readiness</h2> <p>Even if you’re not ready for AI today, here are three things you can do to eventually see significant benefits from AI for the problems you’ve been struggling with.</p> <h3 class="wp-block-heading">1: Start with a manual or semi-automatic approach</h3> <p>If your users are receptive to AI, but you don’t have the data to support the initiative, or you don’t quite understand what problem you’re solving, consider starting with a manual approach.</p> <p>This means you put together a small team (e.g., virtual assistants for non-domain-specific problems) and have them manually execute the tasks while also storing the data from the manual execution.</p> <p>Alternatively, if the workload is extremely high, you can consider automating the task with a less-than-ideal software automation to bring in some level of control to the task.</p> <p>For example, if your virtual assistant is expected to analyze thousands of images to spot a stop sign. But you know that images with specific color distribution will not have a stop sign with 99% certainty. You can develop a simple software script to weed out such images from review, reducing the workload of your virtual assistants. There are many such possibilities to integrate simple software automation before introducing AI.</p> <h4 class="wp-block-heading">Why does this work?</h4> <ul><li>You can easily change the design of your solution until you’re comfortable.</li><li>You can keep changing the type of data you’re collecting</li><li>You can generate high-quality data for machine learning down the road</li><li>You can establish baseline metrics, which you can later compare with an AI-powered solution</li></ul> <h3 class="wp-block-heading">2: Collect data efficiently</h3> <p>If your only problem is the lack of data to develop your AI solution, there are ways to do this efficiently without completely spawning out a full data strategy.</p> <p>I will not get into this in-depth in this article, as you can read my article where I talk about strategies for <a href="https://www.opinosis-analytics.com/blog/machine-learning-training-data/" target="_blank" rel="noreferrer noopener">generating data for your machine learning projects</a>.</p> <h3 class="wp-block-heading">3: Remove adoption fears</h3> <p>If your problem is well understood and you have the necessary data, but users want nothing to do with an automated solution, there’s much work to do on the cultural side of things.</p> <p>You’d need to think about how to get buy-in from users, who may be your employees, customers, or even vendors.</p> <p>The way to approach this is first to ask them what they think about automating specific tasks. If you sense resistance, you’d want to understand their worries and fear. This will give you a sense of what your emphasis would be when you’re trying to “sell” them a solution.</p> <p>If the worry is <a href="https://kavita-ganesan.com/will-ai-kill-your-job/" target="_blank" rel="noreferrer noopener"><em>fear of job loss</em></a>, you can show employees how the nature of their work will change or be simplified with the integration of AI.</p> <p>If the fear is potential rigidity in workflows, you can educate users about how you’re not just developing a solution in the dark but rather co-developing one with them (the users) to ensure that they’re happy with it and it’s truly solving a pain point.</p> <h4 class="wp-block-heading">Why does this work?</h4> <ul><li>By actively removing fears and seeking feedback, you’re fostering collaboration.</li><li>The more collaboration between potential consumers of AI, business stakeholders, designers, and developers, the better the solution and the higher the chances of AI adoption.</li></ul> <h2 class="wp-block-heading">Last Word</h2> <p>As you’ve seen in this article, although your business problem may be a good candidate for AI, you may not be ready to start with AI.</p> <p>There’s a high likelihood that you lack the data for the initiative, may not understand the problem well, and your users may not be ready for an automated solution. The workaround to this is not to start with AI—but to start without it. Solve the foundational issues using simple but effective approaches.</p> <p>That’s all for now.</p> <p></p> <h2 class="wp-block-heading">To Keep Learning From Me:</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li><li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (get 3 free chapters <a href="https://kavita-ganesan.com/the-business-case-for-ai/">here</a>).</li><li><strong><a href="/contact-me">Send me a message</a> </strong>for a more customized roadmap and AI strategy for your business</li></ul> <h2 class="wp-block-heading">Recommened Articles</h2> <ul><li><a href="https://kavita-ganesan.com/will-ai-kill-your-job/">Will AI Really Kill Your Job?</a></li><li><a href="https://kavita-ganesan.com/machine-learning-training-data-generation/">5 Strategies for Generating Data for AI</a></li></ul> <p></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/how-to-improve-ai-outcomes-in-business/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">8000</post-id> </item> <item> <title>15 Common AI Problem Types</title> <link>https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/</link> <comments>https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Fri, 29 Jul 2022 01:48:00 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=6982</guid> <description><![CDATA[One of the problems business leaders face in communicating with their technical counterparts is trying to describe their AI problem. To simplify some of the communication, here are some common AI problem types.]]></description> <content:encoded><![CDATA[<p>One of the problems business leaders face in communicating with their technical counterparts is trying to describe their AI problem. To simplify some of the communication, here are some common AI problem types.</p> <p>Try to map AI opportunities at hand to these common problem types. Note that the problem types often overlap—but that’s ok. The key is to identify problem types that most closely match the task at hand when communicating with your AI and data science experts.</p> <div class="wp-block-uagb-table-of-contents uagb-toc__align-left uagb-toc__columns-2 uagb-block-24899452 " data-scroll= "1" data-offset= "30" style="" > <div class="uagb-toc__wrap"> <div class="uagb-toc__title"> Table Of Contents </div> <div class="uagb-toc__list-wrap"> <ol class="uagb-toc__list"><li class="uagb-toc__list"><a href="#common-ai-problem-types" class="uagb-toc-link__trigger">Common AI Problem Types</a><ul class="uagb-toc__list"><li class="uagb-toc__list"><a href="#1-classification" class="uagb-toc-link__trigger">1. Classification</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#2-regression" class="uagb-toc-link__trigger">2. Regression</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#3-recommendation" class="uagb-toc-link__trigger">3. Recommendation</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#4-search-relevance" class="uagb-toc-link__trigger">4. Search Relevance</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#5-information-extraction-ie" class="uagb-toc-link__trigger">5. Information Extraction (IE)</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#6-text-summarization" class="uagb-toc-link__trigger">6. Text Summarization</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#7-clustering" class="uagb-toc-link__trigger">7. Clustering</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#9-virtual-ai-assistant" class="uagb-toc-link__trigger">9. Virtual AI Assistant</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#10-sentiment-analysis" class="uagb-toc-link__trigger">10. Sentiment Analysis</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#11-object-detection" class="uagb-toc-link__trigger">11. Object Detection</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#12-document-segmentation-problem" class="uagb-toc-link__trigger">12. Document Segmentation Problem</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#13-keyword-extraction" class="uagb-toc-link__trigger">13. Keyword Extraction</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#14-speech-recognition" class="uagb-toc-link__trigger">14. Speech Recognition</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#15-machine-translation" class="uagb-toc-link__trigger">15. Machine Translation</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#summary-of-ai-problem-types" class="uagb-toc-link__trigger">Summary of AI Problem Types</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#other-articles-you-may-be-interested-in" class="uagb-toc-link__trigger">Other Articles You May Be Interested In:</a></li></ul></li><li class="uagb-toc__list"><a href="#keep-learning-succeed-with-ai" class="uagb-toc-link__trigger">Keep Learning & Succeed With AI</a></ul></ol> </div> </div> </div> <p></p> <h2 class="wp-block-heading" id="0-common-ai-problem-types">Common AI Problem Types</h2> <h3 class="wp-block-heading" id="1-1-classification-"><strong>1. Classification </strong></h3> <p>A <strong>classification</strong> problem is about assigning one or more categories to a document, product, person, or image—essentially anything. Examples include:</p> <ul><li>Categorizing incoming support tickets by relevant topics</li> <li><a href="https://www.opinosis-analytics.com/blog/ai-in-manufacturing-examples" data-type="post" data-id="14192">Classifying images of silicon wafers </a>as containing defects or no defects</li></ul> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" class="wp-image-10572" src="https://www.opinosis-analytics.com/wp-content/uploads/2020/07/big_data_strategy_ticket_routing-1024x717.png" alt="Big data strategy" width="536" height="375"><figcaption>Example of support ticket classification</figcaption></figure></div> <h3 class="wp-block-heading" id="2-2-regression-"><strong>2. Regression</strong></h3> <p>A <strong>regression problem</strong> is about estimating <em>numerical</em> values given some input. For example, trying to predict the number of months before a machine needs service given the conditions of the current machine, or predicting how specific drug dosage affects blood pressure. </p> <div class="wp-block-image"><figure class="aligncenter size-full"><img loading="lazy" decoding="async" class="wp-image-14923" src="https://www.opinosis-analytics.com/wp-content/uploads/2021/12/machine-learning-regression.png" alt="" width="500" height="337"><p> </p><figcaption>Predicting a person’s weight given their height—a regression problem. Source: <a href="https://online.stat.psu.edu/stat500/book/export/html/478" target="_blank" rel="noreferrer noopener">stat.psu.edu</a></figcaption></figure></div> <h3 class="wp-block-heading" id="3-3-recommendation-"><strong>3. Recommendation</strong></h3> <p>A <strong>r</strong><strong style="font-size: 1rem;">ecommendation problem </strong><span style="font-size: 1rem;">is about providing personalized content or products to a group of people. </span>Examples include:</p> <ul><li>Product recommendation</li> <li>Recommendations on who to follow</li> <li>Recommendations on jobs to apply for</li> <li>Recommendations on articles to read</li></ul> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" class="wp-image-14926" src="https://www.opinosis-analytics.com/wp-content/uploads/2021/12/topics-to-follow-—-recommendations-1024x655.png" alt="" width="700" height="448"><p> </p><figcaption>Recommendations of topics to follow on Twitter. Source: twitter.com</figcaption></figure></div> <h3 class="wp-block-heading" id="4-4-search-relevance-"><strong>4. Search Relevance</strong></h3> <p>A <strong>search relevance</strong> problem is about <span style="font-size: 1rem;">improving the rankings of search results shown to users.</span> Often search relevance improvement starts with the analysis of search logs to diagnose problems using hard data. Search improvement may or may not require heavy use of machine learning.</p> <h3 class="wp-block-heading" id="5-5-information-extraction-ie"><strong>5. Information Extraction</strong> (IE)</h3> <p>An<strong> information extraction</strong> problem is about extracting specific information from large volumes of text data. One of the goals of information extraction is to fill templates using data extracted from raw text. Examples include:</p> <ul><li>Extracting patient symptoms from large volumes of clinical notes</li> <li>Extracting pertinent information from large volumes of legal case files</li> <li>Pre-populating a candidate application form / database by extracting pertinent information from resumes</li></ul> <h3 class="wp-block-heading" id="6-6-text-summarization-"><strong>6. Text Summarization</strong></h3> <p><strong>Text summarization </strong>is about creating an accurate synopsis of a longer document or a set of documents. </p> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" class="wp-image-14941" src="https://www.opinosis-analytics.com/wp-content/uploads/2021/12/review-summarization-1024x609.png" alt="" width="599" height="356"><p> </p><figcaption>An example of review summarization</figcaption></figure></div> <h3 class="wp-block-heading" id="7-7-clustering-"><strong>7. Clustering</strong></h3> <p><strong>Clustering</strong> is about grouping people, content, documents, topics and etc based on some logical structure—for example, grouping customers by their purchase behavior. </p> <p>More generally, clustering divides data points into a number of fixed (or dynamic) groups such that the data points in one group are more similar to each other than data points in other groups. </p> <h3 class="wp-block-heading" id="8-9-virtual-ai-assistant-"><strong>9. Virtual AI Assistant</strong></h3> <p><strong>Virtual AI Assistant </strong>is used for having short conversations with humans to complete simple tasks. Examples include:</p> <ul><li>Providing answers to common customer questions without human involvement</li> <li>Using text messages to check bank balances or to get a refund</li></ul> <p>Alexa and Siri are examples of virtual AI assistants.</p> <h3 class="wp-block-heading" id="9-10-sentiment-analysis-"><strong>10. Sentiment Analysis</strong></h3> <p><strong style="font-size: 1rem;"><a href="https://kavita-ganesan.com/sentiment-analysis-business-applications/#.Y0C-Z-zMJhE">Sentiment Analysis</a> </strong><span style="font-size: 1rem;">is about discovering emotions in text data such as user reviews, social media comments, and surveys. For example, automatically detecting customer sentiment in social media channels after a new product release.</span> Sentiment analysis can even be applied to images for understanding emotions from facial expressions.</p> <h3 class="wp-block-heading" id="10-11-object-detection-"><strong>11. Object Detection</strong></h3> <p><strong style="font-size: 1rem;">Object detection problem </strong><span style="font-size: 1rem;">is about discovering specific objects such as humans, buildings, or cars in digital images and videos.</span></p> <div class="wp-block-image"><figure class="aligncenter size-large"><img loading="lazy" decoding="async" class="wp-image-14917" src="https://www.opinosis-analytics.com/wp-content/uploads/2021/12/object-detection-1024x538.jpeg" alt="" width="1024" height="538"><p> </p><figcaption>Example of object detection—automatically detecting people, traffic lights, personal accessories, and vehicles given an image. Source: <a href="https://infotech.report/guest-articles/real-time-object-detection-in-video-with-intro-to-yolo-v3" target="_blank" rel="noreferrer noopener">infotech.report</a></figcaption></figure></div> <h3 class="wp-block-heading" id="11-12-document-segmentation-problem-"><strong>12. Document Segmentation Problem</strong></h3> <p><strong>Document segmentation</strong> is about trying to subdivide documents into meaningful parts. For example, segmenting unstructured clinical texts to extract their past medical history and family history. </p> <div class="wp-block-image"><figure class="aligncenter size-large"><img loading="lazy" decoding="async" class="wp-image-14915" src="https://www.opinosis-analytics.com/wp-content/uploads/2021/12/document-segmentation-1024x677.png" alt="" width="1024" height="677"><p> </p><figcaption>Example segmentation of a clinical record</figcaption></figure></div> <h3 class="wp-block-heading" id="12-13-keyword-extraction-"><strong>13. Keyword Extraction</strong></h3> <p><strong>Keyword extraction</strong> <span style="font-size: 1rem;">is about identifying terms that best describe the subject of a document—for example, extracting keywords from large volumes of legal documents to understand the themes of discussion.</span></p> <p>While there are many keyword extraction tools readily available (including open-source tools), you’d need to ensure that these work on your data. Often, keyword extraction tools are best customized or custom-developed.</p> <h3 class="wp-block-heading" id="13-14-speech-recognition-"><strong>14. Speech Recognition</strong></h3> <p>Speech recognition, also known as speech-to-text (STT) or automatic speech recognition (ASR), is about having a computer program understand and transform spoken language into a written format (or text).</p> <p>Speech recognition is often used to complete downstream tasks. For example, speech recognition is used behind the scenes to surface relevant search results when you use Google voice search. Specifically, your speech is translated into a human-readable format, and that generated text is used to surface relevant search results.</p> <p>Many vendors offer speech recognition solutions, and therefore, speech recognition systems rarely need to be developed from scratch. Of course, these systems will benefit from customization for the target data.</p> <h3 class="wp-block-heading" id="14-15-machine-translation-"><strong>15. Machine Translation</strong></h3> <p><a href="https://www.opinosis-analytics.com/blog/nlp-applications/#3-machine-translation">Machine translation</a> is the automatic software translation of text from one language to another. For example, translating English sentences into German with reasonable accuracy. Machine translation programs rarely need to be developed from scratch but may benefit from customization.</p> <p>Machine translation is used for many purposes, including:</p> <ul><li>Localizing website text for a particular country</li> <li>Customer support conversations across countries</li> <li>Understanding documents written in a different language</li> <li> </li></ul> <h3 class="wp-block-heading" id="15-summary-of-ai-problem-types">Summary of AI Problem Types</h3> <p>In this short guide, we discussed 15 common AI problems types—that often overlap. For example, you can apply a classification approach for sentiment analysis. However, the key is to identify the problem type that best fits the task at hand. It doesn’t have to be 100% accurate—it’s just semantics. You can continually refine these definitions with the help of your AI experts.</p> <p></p> <p></p> <div class="inherit-container-width wp-block-group has-ast-global-color-4-background-color has-background is-layout-constrained wp-block-group-is-layout-constrained" style="border-style:dotted;border-width:1px;padding-top:10px;padding-right:10px;padding-bottom:10px;padding-left:10px"><div class="wp-block-group__inner-container"><h2 class="wp-block-heading">Keep Learning & Succeed With AI</h2> <ul><li><strong>Join my <a href="https://kavita-ganesan.com/blog">AI Integrated newsletter</a>, </strong>which clears the AI confusion and teaches you how to successfully integrate AI to achieve profitability and growth in your business.</li> <li><strong>Read <a href="http://amazon.aibusinesscasebook.com/?utm_source=newsletter&utm_medium=email&utm_campaign=why_starting_without_ai_improves_outcomes_even_if_ai_is_your_goal&utm_term=2022-08-19" target="_blank" rel="noreferrer noopener">The Business Case for AI</a></strong> to learn applications, strategies, and best practices to be successful with AI (select companies using the book: government agencies, automakers like Mercedes Benz, beverage makers, and e-commerce companies such as Flipkart).</li> <li><a href="https://kavita-ganesan.com/work-with-me/"><strong>Work directly with me</strong></a> to improve AI understanding in your organization, accelerate AI strategy development and get meaningful outcomes from every AI initiative.</li></ul></div></div> <p></p> <h3 class="wp-block-heading">Other Articles You May Be Interested In:</h3> <ul><li><a href="https://kavita-ganesan.com/tay-twitter-bot/#.Y1nTYOzMJhE">What Happened to Tay the Twitter Bot that Turned Racist?</a></li> <li><a href="https://kavita-ganesan.com/ai-leadership-mistakes/">3 Mistakes Leaders Should Avoid When Integrating AI</a></li></ul>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/common-ai-problem-types-cheat-sheet/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">6982</post-id> </item> <item> <title>4 Business AI Predictions for 2022-2023</title> <link>https://kavita-ganesan.com/ai-predictions/</link> <comments>https://kavita-ganesan.com/ai-predictions/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Tue, 28 Jun 2022 23:59:21 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Strategy]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=7159</guid> <description><![CDATA[Having worked with numerous clients, big and small in the integration of AI, here are 4 Business AI predictions in 2022 and beyond.]]></description> <content:encoded><![CDATA[<p>AI as a field, especially in the context of real-world applications, has been progressing at a rapid pace. This has been further accelerated by the onset of the COVID-19 pandemic. In fact, AI was found to be the<a href="https://www.tvtechnology.com/news/artificial-intelligence-was-the-most-discussed-technology-of-2021"> most discussed technology</a> in 2021. Having worked with numerous clients, big and small, in the integration of AI, here are 4 Business AI predictions in 2022 and beyond.</p> <h2 class="wp-block-heading"><strong>#1 Many more “deployed” models</strong></h2> <p>In the recent past, businesses have had trouble operationalizing models and have not seen the value in many of their AI initiatives. In fact, <a href="https://www.gartner.com/en/newsroom/press-releases/2020-10-19-gartner-identifies-the-top-strategic-technology-trends-for-2021">Gartner’s research </a>shows that <em>only</em> 53% of AI initiatives make it from prototype to production.</p> <p>However, with the recent growth in the number of ML deployment platforms, low-code and no-code AI development services, and AI vendors providing out-of-the-box AI solutions, such as speech recognition, sentiment analysis, and ticket routing, we will start witnessing many more real-world applications of AI.</p> <h2 class="wp-block-heading"><strong>#2 The rise of problem-focused practitioners</strong></h2> <p>While the focus of AI as a field has been strongly techniques-focused, business practitioners are slowly beginning to realize that the latest and greatest techniques <em>may not necessarily work</em> from a practical standpoint for many use cases. There is a fundamental difference between techniques that are still in “research mode” versus those that have been tried and tested.</p> <p>Even though in the past, data scientists have taken pride in using the most sophisticated techniques to demonstrate expertise, data scientists are becoming more <em>problem-focused.</em> They’re adopting simpler techniques that will have a higher chance of success for real-world use cases. This change in thinking will improve the outcomes of many AI initiatives.</p> <h2 class="wp-block-heading"><strong>#3 </strong><strong><em>Accountable AI</em></strong><strong> will gain steam</strong></h2> <p>With all the Facebook drama and regulations around AI still being limited and “in the talks”, more and more businesses are becoming aware of the problems with algorithms. Unless algorithms are used responsibly with downstream and long-term impact in mind, it’s clear that they can do significant damage. To that end, I’m seeing many data scientists and leaders talk about the ethics and implications of algorithms.</p> <p>I believe that informal conversations around AI ethics are just the beginning. Some of these discussions will turn into action where businesses will start having their <em>own committees</em> to vet AI systems rigorously. Some will even study potential societal impact before the release of specific technologies—regardless of regulations. Regulations will only add another layer of oversight, especially for companies that have yet to take AI ethics and accountability seriously.</p> <h2 class="wp-block-heading"><strong>#4 </strong><strong><em>Underserved </em></strong><strong>industries will start adopting AI</strong></h2> <p>AI has largely been a winning tool for large tech companies. But the stress caused by the COVID-19 pandemic, such as the shrinking labor force participation, workers not wanting to work on the front lines, social distancing requirements and others have forced many companies that used to rely heavily on the availability of workers to <em>rethink </em>their business models.</p> <p>From allowing employees to work remotely, to automating away jobs that no human worker wants to do, are all options on the table for serious consideration. As part of this, AI will be a critical player in changing businesses forever. Hospitals, manufacturers, and restaurant chains will all be at the crossroads of technology transformations.</p> <h1 class="wp-block-heading"><strong>Summary</strong></h1> <p>While AI within business applications was a new concept several years ago, as you can see from these predictions that it’s becoming more mainstream and achievable. The growth of AI deployment and development platforms, mindset changes, and stress caused by the COVID-19 pandemic will all be true catalysts in making AI a reality for businesses.</p> <p>The post <a href="https://www.opinosis-analytics.com/blog/2022-2023-business-ai-predictions" rel="nofollow">4 Business AI Predictions for 2022-2023</a> appeared first on <a href="https://www.opinosis-analytics.com" rel="nofollow">Opinosis Analytics</a>.</p> <p><a href="https://www.opinosis-analytics.com/blog/2022-2023-business-ai-predictions" target="_blank" rel="noopener">4 Business AI Predictions for 2022-2023</a></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/ai-predictions/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">7159</post-id> </item> <item> <title>What is AI? And what does it mean for me and the world?</title> <link>https://kavita-ganesan.com/what-is-ai-and-what-does-it-mean-for-me-and-the-world/</link> <comments>https://kavita-ganesan.com/what-is-ai-and-what-does-it-mean-for-me-and-the-world/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Tue, 24 May 2022 14:20:43 +0000</pubDate> <category><![CDATA[Podcast Interview]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=7453</guid> <description><![CDATA[What is AI? And what does it mean for me and the world? A Podcast Interview: Kavita Ganesan & David Gee. ]]></description> <content:encoded><![CDATA[<p><em>Listen to my interview with the <strong><a href="https://www.audible.com/pd/Insatiably-Curious-Podcast-Podcast/B08JJN4W7Z">Insatiably Curious Podcast</a> </strong>host, David Gee.</em> <em>Transcript of the podcast is as below.</em></p> <p>00:07</p> <p>Welcome to the <strong>Insatiably Curious Podcast</strong>, where we invite lifelong learners to join us on a personal and professional journey. Now here to inform, entertain and enlighten, while always keeping it interesting from our nation’s capital. It’s your host, David Gee.</p> <p><strong>David Gee </strong>00:26</p> <p>Joining us today so glad to have her, Kavita Ganesan… She is the author of <strong><em>“The Business Case for AI: A Leader’s Guide to AI Strategies, Best Practices & Real-World Applications.”</em></strong> Welcome to the show, Kavita.</p> <p><strong>Kavita Ganesan </strong>00:40</p> <p>Yes, thank you for having me, David.</p> <p><strong>David Gee </strong>00:41</p> <p>So one of the things when we have broad complicated topics, and I certainly think this qualifies… I like to kind of set the landscape or the ground rules a little bit to make sure we’re talking about the same thing or that we even know, in some cases, what we are talking about it.</p> <p>So we’re talking about Artificial Intelligence, AI, how it informs the kinds of decisions we can make it work. But why don’t you give us the layman’s definition of what it is we’re talking about, and maybe second to that the difference between how you might use it in your computer science lab in an academic or research setting and how I might use it at my office or IBM or something like that.</p> <p><strong>Kavita Ganesan </strong>01:33</p> <p>Yes, so AI is all about trying to mimic human decision making, within a computer. And using software programs. And the way AI systems today work is by learning from data. So let’s take a credit card. Let’s say you’re trying to detect fraudulent transactions, credit card transactions. So the way that AI systems today learn is by looking at 1000s of different examples of what makes a transaction fraudulent, or non-fraudulent. And then it mines the patterns [phonetic 02:09] from that. And then the next time it sees a new transaction, it then decides, hey, this seems suspicious. So it might be fraudulent. So that’s how systems today learn. But historically, they have been very rules based. And in the future, it may not even be data dependent anymore.</p> <p>So right now it’s highly data dependent. And from a business perspective, AI systems are very good in improving efficiency of workflows. So let’s say, right now, your customer service agent is manually routing tickets to the appropriate teams to get things resolved. So if you put an AI system there, it can do that very efficiently, 24 hours a day, and maybe even more accurately than your human agent.</p> <p>But from a research perspective, in a research lab, that’s not how we look at things, we are not looking at what benefits that can be to a business. So we are thinking about how can we get better? How can technique-2 become better than technique-1. So we are always looking at incremental improvements, either through better data or better techniques. So we’re always trying to beat the state of the art. So it’s kind of different, what happens in the research world versus what’s happening in the business world. And the business we have to be completely focused on the applications of it, how it’s going to help us.</p> <p><strong>David Gee </strong>03:40</p> <p>There’s probably not a day that goes by that I don’t come across AI and in Washington Post, New York Times, Wall Street Journal, you know, whatever your source of news is. It’s certainly on the minds of business leaders today, is it currently in the state that we use it today as it is a kind of the purview of large corporations, IBM’s and Googles and in Tesla’s, or are you seeing the use of AI filtered down to more medium sized businesses or even in some cases, small businesses?</p> <p><strong>Kavita Ganesan </strong>04:24</p> <p>So there are two groups that I see, that have really embraced AI, and that’s the large corporations. And that’s specifically on the tech sector, not in other industries, and startups…. This AI startup are very good using AI and they’re very efficient because they’re small, they’re nimble, they know how to adopt new technologies and just run with it. And then the large corporations on the tech side, they are already very AI driven. Their infrastructure is set up for AI. So these two groups are running fast with AI but I think that’s the group in between the midsize businesses, they are thinking about AI, but they just don’t know how to get started. Because there’s so much of confusion in the media, like what’s happening in research gets overstated as the current capabilities.</p> <p><strong>David Gee </strong>05:16</p> <p>In your book, you talk about some of the myths, conceptions and the myths around AI that we should all be aware of… Why don’t you specifically outline a couple of them or predominant ones?</p> <p><strong>Kavita Ganesan </strong>05:30</p> <p>Yes, sure. So the first myth that people think is that they need to use the latest and greatest technique that the media is talking about. But really, the latest and greatest techniques are very still in the r&d phase. And they may not fit in within your infrastructure, you may have an old infrastructure that can maybe take in basic computer algorithms, not something sophisticated that needs GPUs, and TPUs. And you have to keep in mind that AI has been there since the 1940s. So the techniques are already very old, it’s only become popular now because of the computation power that we have. So using any one of those techniques that can benefit your business, is a good technique. So not to worry about what’s being discussed in the media.</p> <p><strong>David Gee </strong>06:20</p> <p>I don’t necessarily want to make this a history lesson, but you just told me something I absolutely was not aware of. And that is 7 decades give or take of AI. So without supercomputers and the computational power that we have today, how was AI used in the 1940s, the 1950s, I’m fascinated by that.</p> <p><strong>Kavita Ganesan </strong>06:44</p> <p>Yes, so back in the days, it was like heavily rules base, you have to encode human knowledge in the form of specific rules. And they also had started neural networks research long ago, but that research stopped because insufficient computation power. And then he picked up again, I think, around 2011, when big data became a thing, then we had lots of faster computation power, and then it just accelerated from that. And neural networks now has become deep learning. And that’s a big field of study now.</p> <p><strong>David Gee </strong>07:19</p> <p>You talk again, in your book about the 5 basic pillars of AI preparation and AI landscape, why don’t you go into a little bit of those, if you would, please.</p> <p><strong>Kavita Ganesan </strong>07:32</p> <p>Yes, so given how AI systems today learn, like they heavily dependent on data. So if you’re not already collecting data, or if your processes are still paper based, then data infrastructure is a big pillar in your AI preparation. And you don’t have to start off fancy, you just have to make sure that you’re doing the basics to start with, like, if you’re using paper processes, why not shift to Excel. If you’re not collecting data, think about where, which are the areas that AI could benefit, and start collecting data in those areas. So that’s one huge pillar, the data infrastructure pillar, then there is the cultural pillar. That’s like addressing some of the fears around AI, because a good percentage of Americans are fearful of AI, and a research has actually shown that [unintelligible 08:36] that in my book. So you want to put this fears to rest so that companies can actually start looking into the benefits of AI, as opposed to fearing AI. Then there’s also understanding what’s development around AI? Like, that’s also a cultural element. It’s very experimental. It’s very iterative. So you need to establish those cultural elements. So those are the two big pillars:</p> <ol type="1"><li>Data</li></ol> <p>and</p> <ul><li>Culture.</li></ul> <p><strong>David Gee </strong>09:09</p> <p>Follow up on that fear angle, oftentimes just kind of the way that humans are wired, we fear the unknown, whatever that is. Is this what’s it at play here? Is it simply that most people don’t know enough about AI to feel comfortable with it? Or is it something kind of spookier where they envision giant super computers and robots and everybody or things like that, you know, kind of taking over the world with information and you know, brains mimicking my brain and so on. And so, you know, what, what is it that that, you know, I know you’re not a psychologist, but in your take, what is it that people fear about the subject of AI?</p> <p><strong>Kavita Ganesan </strong>10:00</p> <p>So, there are some celebrities out there that are spreading this type of information that it will take of our jobs and later humanity, it can be used for bad things. And some of it is true, like AI can be used unethically. But taking over humans, that’s untrue. Because AI systems today don’t have the common sense reasoning as humans do, they can’t read body language, they don’t natively understand emotions, they can’t read between the lines, and just connect the dots like humans can. But on very specific tasks, AI systems can be very effective, and can even surpass human accuracy.</p> <p>So the misunderstanding is that these AI systems can become conscious like humans, and then start making decisions that can overpower humans, I’ve seen that is a real fear, really. And that actually even came up and I was trying to hire one of my narrators for my book. And she said, Hey, I was afraid to audition for this book, because there’s so much fear about AI in our community. So, it’s widespread.</p> <p><strong>David Gee </strong>11:14</p> <p>Interesting. Yeah, I can think of a couple of Hollywood movies where the machines are plotting… Ex Machina is actually one of the [unintelligible 11:23] a machine plots to take over the mind and world of the founder or the inventor of the machine. So yeah, Hollywood has definitely taken that theme. And I’m sure some media outlets as well. You talk about back to the book, you talk about what you call vanity AI, you know, anything that’s kind of in the zeitgeist or that becomes a kind of a thing will be social media or anything else. There are companies that will just kind of glom on to that not in a real authentic sense, but just to say, hey, yes, guess what, yes, we do AI too… You talk about vanity AI?</p> <p><strong>Kavita Ganesan </strong>12:11</p> <p>Yes, Vanity AI happens a lot. And a lot within large corporations, especially, because there’s a rush to adopt AI. And as that trickles down the ladder, people think that they just need to use AI, and the way companies start using AI today is just by looking at data, seeing what data is available, and then coming up with problems based on data. But the problem with that approach is that it tells you nothing about the inefficiencies within the company, like what problem really, is it solving? What pain point is it addressing? So because of that the AI projects don’t necessarily solve real business challenges. And then executives don’t see the value from it. And then they start distrusting AI as a whole, that it’s useless. It’s just hype. But really, there’s not enough planning around it for what problems are we applying AI.</p> <p><strong>David Gee </strong>13:12</p> <p>So, I’m in the content creation, marketing messaging business, and I use data all the time to inform my content, whether it be page views, or how long people engage with content, you know, downloads, you know, all those kinds of things. We have lots of ways to measure the way that people interact with our marketing with our messaging, is there do you draw a distinct line between what I would call data driven decision making and AI? Am I using AI a little bit when I make those content decisions? Or is it just something completely different?</p> <p><strong>Kavita Ganesan </strong>13:57</p> <p>Yes, that’s a great question. So AI can serve two grand purposes.</p> <ol type="1"><li>One is to improve efficiency within businesses.</li></ol> <ul><li>And one is to help make better decisions.</li></ul> <p>And if you look at our data, we have a lot of structured data, things that fit neatly in an Excel spreadsheet. And then we have lots of unstructured data, like all the Twitter comments, all the documents within your company, your customer support conversations, all of that is completely unstructured. So when you’re trying to extract data-driven insights, you can’t just use your structured data, which I refer to as simple data analytics, but to make deeper decisions, like what are your customers complaining about? What’s your top wish list? You need that unstructured data and you can’t really aggregate unstructured data like you do structured data, you need some form of AI on that, and specifically, a branch of AI called natural language processing that tries to make sense of all that text data, and then extract key elements that you can later make sense of and inform decisions like, what are my customers’ top points, wishlist and so on. So there’s a lot of opportunity there.</p> <p><strong>David Gee </strong>15:16</p> <p>Well, certainly that… I mean, regardless of how you achieve, it was certainly from a customer service perspective, a marketing perspective, the more we know about our audience, our prospects, our consumers, the better we’re going to be and the bigger competitive advantage we have, right? And it’s not just,</p> <p><strong>Kavita Ganesan </strong>15:38</p> <p>Yes, that’s right.</p> <p><strong>David Gee </strong>15:38</p> <p>Yes. So when you first came to University of Southern California to get a master’s in computer science, were you hearing about AI from your first day in school was AI, a big thing in the halls of computer science classes a long… a while ago.</p> <p><strong>Kavita Ganesan </strong>16:00</p> <p>So USC is a special breed, they were so much into AI, even when I joined. So, that’s how I got exposed to this whole field. I took one natural language processing class and the professor. And I was hooked. I liked it. So then I went on to doing my PhD in the field, and then became a data scientist. But within USC, AI was a big thing. Within academic institutions, AI has been a big thing since I guess long ago.</p> <p><strong>David Gee </strong>16:32</p> <p>And you’ve been at the consulting game for 15 years, give or take and working with Fortune 500 companies, as well as some smaller organizations, when you first you know… When you got your freshly minted degrees and kind of went out in the business world, were people conversing about AI? Did you have to do a lot of educating. Tell me about the landscape when you first started.</p> <p><strong>Kavita Ganesan </strong>17:00</p> <p>So when I first started, AI was not known in the industry. So I had to become a software engineer, because there was no prospect for AI right after I graduated with my master’s. So I did my PhD thinking that I’ll go into a research lab to maybe use some AI. But around 2013 is where data science started to really become a thing. So then I jumped on to the whole data science field where I could apply AI. So it’s only around 2013, 2011, where AI became a thing in the industry.</p> <p><strong>David Gee </strong>17:46</p> <p>We spent a little bit of time today talking about both the past and the future, as you look into to the future and to your crystal ball. What do you think is and conversely, is not appropriate to expect from Ai?</p> <p><strong>Kavita Ganesan </strong>18:04</p> <p>I think what we can expect is as the understanding around AI improves, I think all the midsize businesses are going to start using AI the right way. And also the non tech companies are going to start embracing AI and start seeing value from it. But in the next few years, don’t expect a conscious bot to be walking around. That’s not going to happen in the next few years. And maybe not even this whole decade. So these are the two things that are going to happen. Business are going to pick up on AI a lot more. And research is going to improve around trying to mimic like human reasoning within an AI system, but not to the point of it becoming conscious, really.</p> <p><strong>David Gee </strong>18:51</p> <p>So you do see it filtering down into the mainstream and into smaller companies, though, is that correct?</p> <p><strong>Kavita Ganesan </strong>18:57</p> <p>Definitely. Yes.</p> <p><strong>David Gee </strong>18:59</p> <p>Anything that you’re concerned about as a data scientist, as a software engineer, as someone who’s spent so much time around AI? Is there a potential? I mean, we don’t know, you know, the capacity for human beings sometimes. But is there something that you’re even as knowledgeable as you are that you’re a little fearful of?</p> <p><strong>Kavita Ganesan </strong>19:22</p> <p>So my biggest fear is one bias in data. So that can propagate through AI systems because AI systems, learn from data and bias has been shown to be very prevalent, like in facial recognition systems. So in fact, some of the big tech companies stop using facial recognition systems for law enforcement, I think?</p> <p><strong>David Gee </strong>19:47</p> <p>Right.</p> <p><strong>Kavita Ganesan </strong>19:48</p> <p>Yes. So that’s one big problem, the bias in data and the second problem is how people are using AI like we saw with Deep fakes. The recent news about Deep fakes on how people can make you say things that you didn’t really say. And then make that to be reality. So the applications of AI really needs to be regulated. Otherwise, it’s going to be used in unethical ways.</p> <p><strong>David Gee </strong>20:15</p> <p>Wow. Yes. I mean, talking about big tech and regulation. I live in Washington, DC. And that’s obviously, you know– the pages of The Washington Post all the time about how we regulate these companies and social media and that could be a whole another conversation. So thank you so much for joining us. I’ve really enjoyed it.</p> <p><strong>Kavita Ganesan </strong>20:36</p> <p>Yeah, thank you for having me, David.</p> <p><strong>David Gee </strong>20:38</p> <p>Kavita Ganesan, author of <strong><em>“<a href="https://www.amazon.com/Business-Case-Strategies-Real-World-Applications-ebook/dp/B09TRS55K8?crid=1UX93S9MHCFXF&keywords=the+business+case+for+ai&qid=1649883890&sprefix=the+business+case+for+ai,aps,136&sr=8-1&linkCode=sl1&tag=kavganbook-20&linkId=ce9c7b637e704bd00a06f421c7a17016&language=en_US&ref_=as_li_ss_tl">The Business Case for AI: A Leader’s Guide to AI Strategies, Best Practices & Real-World Applications.</a>” </em></strong>Thanks again to her and thanks to you for joining us as well. I’m David Gee. We’ll see you next time. So long, everyone.</p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/what-is-ai-and-what-does-it-mean-for-me-and-the-world/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">7453</post-id> </item> <item> <title>Will AI Kill Your Job?</title> <link>https://kavita-ganesan.com/will-ai-kill-your-job/</link> <comments>https://kavita-ganesan.com/will-ai-kill-your-job/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Fri, 04 Feb 2022 18:27:08 +0000</pubDate> <category><![CDATA[AI For Business]]></category> <category><![CDATA[AI Mindset]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=6806</guid> <description><![CDATA[Will AI *really* steal our jobs? This article performs some reality checks on this topic. You may find some of the thought process counterintuitive to what the media makes you believe.]]></description> <content:encoded><![CDATA[<p><em><strong>Oh, what will happen to our jobs when my company starts using AI?</strong></em></p> <p>This is such a popular question that comes up when I give talks—even to a pseudo technical audience.</p> <p>Not surprisingly, this question will land sarcastic smiles on the faces of some data scientists and AI experts; you may even catch some eyes rolling. </p> <p><em>But it’s a valid question. </em></p> <p>Don’t forget, we had a US political candidate run for the presidency on the theme that <a href="https://www.nytimes.com/2019/11/14/opinion/andrew-yang-jobs.html" target="_blank" rel="noopener">automation is robbing people of good jobs</a> and we should thus be compensated for the crisis. Popular figures take it to extreme levels and even state that <a href="https://www.independent.co.uk/life-style/gadgets-and-tech/news/elon-musk-artificial-intelligence-ai-singularity-a9640196.html" target="_blank" rel="noopener">AI will take over the human race</a> very soon.</p> <p>With all this noise around automation and AI, of course, this will send shivers down people’s spines; A technology they don’t quite understand is trying to <em>pull the rug from under their feet</em>. Worse still, this technology has the potential of destroying them, according to some.</p> <p>The fear is real.</p> <p>According to a<a href="https://governanceai.github.io/US-Public-Opinion-Report-Jan-2019/public-opinion-on-ai-governance.html." target="_blank" rel="noopener"> survey</a> conducted by Oxford University’s Center for the Governance of AI, Americans fear a future where AI becomes too intelligent. When people were asked what kind of impact machine intelligence would have on humanity, 34 percent thought it would be negative, with 12 percent leaning towards <em>human extinction</em>! </p> <p>This begs the question of, <em>Will, AI *really* replace our jobs and later us?</em> </p> <p>The answer is complicated. There are several parts to this story.</p> <h1 class="wp-block-heading" id="why-ai-will-steal-some-of-our-jobs">Why AI <em>will</em> steal some of our jobs</h1> <p>To recap, the goal of AI is to mimic human intelligence and decision-making within a computer, leveraging computer algorithms. Given the computing power, learning algorithms, and data we have today, there are many repetitive tasks done by humans that can be offloaded to AI. At least, AI can assist humans in completing those tasks.</p> <p>For example, AI systems can be taught to handle simple customer support requests, detect defects on manufactured products, and even investigate the possibility of a fraud event. Once these AI systems are trained, they can perform these tasks more efficiently, quicker, and sometimes even with higher accuracy. So, of course, specific jobs, especially those decision-making jobs that are also labor-intensive, can be “replaced”. But not in the way you think.</p> <p>Remember that when AI systems fail or make mistakes, you still need humans to rectify the issues. You also need humans in the loop to provide quality assurance–essentially, become the supervisor of these AI systems. Further, as AI systems today rely heavily on data to learn how to perform tasks, you still need humans in the loop to generate high-quality data. More importantly, there’ll be functions around the “stolen” job that no AI is capable of doing single-handedly, such as reporting progress, finding areas of improvement, managing team members, etc.</p> <p>So eventually, while tasks that used to take a whole team may now be augmented with AI, the roles within your team can change. The humans may become acting supervisors, AI trainers, and quality assurance managers.</p> <p>In the end, teams could shrink. But teams may also stay the same for years, while their roles change. In some cases, teams may grow bigger due to the demand for other related roles. Nevertheless, humans play a critical part in AI-augmented workflows.</p> <h1 class="wp-block-heading" id="why-ai-will-not-replace-us-humans">Why AI <em>will not</em> replace us humans</h1> <p>AI systems are task-oriented machines. If you train these machines to perform a series of tasks, they very well will. For example, if you teach a machine learning algorithm to detect breast cancer by asking it to read thousands of mammogram images, it could very well do it like a radiologist. At the same time, the same AI system will not be able to detect any other types of cancer, unlike a radiologist who could detect other related cancers.</p> <p>We must come to terms that AI systems today lack commonsense reasoning, the ability to effectively build upon knowledge, read between the lines, and use logic as humans do. While humans are born with such innate abilities, AI systems today rely heavily on <em>patterns</em> in data.</p> <p>Even the famed <a href="https://kavita-ganesan.com/what-is-gpt-3/"> GPT-3</a>, the massive language prediction model that seems to do everything, including completing computer source code, answering questions, predicting sentiment, writing articles, etc, gets stumped when you play with its “commonsense abilities”. When you trick it with illogical questions, it produces illogical answers. That’s because it’s trained to learn <em>obvious</em> patterns from huge volumes of data, not conceptualized knowledge, which is more abstract.</p> <p>For example, if you ask <a href="https://kavita-ganesan.com/what-is-gpt-3/">GPT-3</a>, “How many eyes does your foot have?”, it will answer, “Your <strong>foot</strong> has <strong>two</strong> <strong>eyes</strong>” (<a href="https://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html" target="_blank" rel="noopener">see examples here</a>). On the contrary, a two-year-old will quickly tell you that feet don’t have eyes!</p> <div class="wp-block-image"><figure class="aligncenter size-large"><a href="https://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html" target="_blank" rel="noopener"><img loading="lazy" decoding="async" class="wp-image-15597" src="https://www.opinosis-analytics.com/wp-content/uploads/2022/01/question-and-answer-with-GPT-3-1024x437.png" alt="" width="1024" height="437"></a><figcaption><strong>Question and answers with GPT-3</strong> (Source: <a href="https://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html" target="_blank" rel="noopener">lacker.io</a>)</figcaption></figure></div> <p>My point here is that while AI systems can perform complex tasks, at present, they’re very much task-oriented machines. AI is a long way from performing humanlike reasoning in a general sense. If one of the more sophisticated AI systems can’t tell you that feet don’t have eyes, do you really think AI robots can plan a complete takeover of the human race? General humanlike reasoning technology just isn’t there yet due to its complexity and it may be so for several decades to come. Nonetheless, this does not diminish the fact that task-oriented AI systems can already enhance human lives in so many ways.</p> <h1 class="wp-block-heading" id="we-need-ai-and-automation-more-than-ever">We <em>need</em> AI and automation more than ever</h1> <p>As we’ve all seen with the COVID-19 pandemic, a 100% human workforce is risky and may no longer be suitable for the world we live in today. Workers resigned in great numbers to either care for their homeschooling kids, fear of the virus itself, or at the realization that many of the jobs were not worth the risk. This has resulted in a <em>shrinking</em> labor force participation which has already been happening over the last several decades due to the aging population and other factors.</p> <div width="100%" align="center"><iframe loading="lazy" align="center" src="https://d3fy651gv2fhd3.cloudfront.net/embed/?s=unitedstalabforparra&v=202201071504V20200908&d1=19970202&type=type=column&h=300&w=600" height="300" width="600" frameborder="0" scrolling="no"></iframe><br>source: <a href="https://tradingeconomics.com/united-states/labor-force-participation-rate">tradingeconomics.com</a></div> <p class="has-text-align-center"><strong>US Labor force participation over the years</strong></p> <p>This means that there will be many unfilled jobs. It could also mean that many businesses won’t survive without alternatives. If enough people aren’t willing to work in roles that were fully filled just a decade ago, companies need to compensate.</p> <p>Automation will become necessary.</p> <p>We’re already seeing much of this play out in pharmacies and grocery stores throughout the United States, where self-checkout machines, heavily software powered, have become more of a norm after the COVID-19 pandemic. It’s just the beginning. As more and more businesses embrace the digital world and with fewer people wanting to work in roles that earlier generations were comfortable working in, there’ll be a lot more need for software-powered automation. Whether companies use AI, RPA, or simple software automation, all of that is context-dependent.</p> <h2 class="wp-block-heading" id="final-word">Final Word</h2> <p>So, in summary, to answer the question of whether jobs will be lost because of AI, the answer is—maybe. But what’s more certain is that the nature of certain jobs will <em>gradually</em> change. Instead of doing everything manually, we’ll slowly be assisted by AI-powered tools. We may also become the quality assurance managers and data generators for AI-powered systems. And as for the question of AI takeover? Well, that’s probably not even a question for this decade—not until we get a hopeful glimpse into the study of AI as it relates to <a href="https://www.quantamagazine.org/common-sense-comes-to-computers-20200430/" target="_blank" rel="noopener">commonsense reasoning</a>.</p> <h1 class="wp-block-heading" id="references">References:</h1> <ul><li><em>US Bureau of Labor Statistics Latest Numbers. Accessed January 28, 2022. <a href="https://www.bls.gov/" target="_blank" rel="noopener">https://www.bls.gov/</a>.</em></li> <li><em>Pavlus, John. 2020. “Common Sense Comes Closer to Computers.” Quanta Magazine. <a href="https://www.quantamagazine.org/common-sense-comes-to-computers-20200430/." target="_blank" rel="noopener">https://www.quantamagazine.org/common-sense-comes-to-computers-20200430/.</a></em></li> <li><em>“Elon Musk claims AI will overtake humans ‘in less than five years.’” 2020. The Independent. <a href="https://www.independent.co.uk/life-style/gadgets-and-tech/news/elon-musk-artificial-intelligence-ai-singularity-a9640196.html" target="_blank" rel="noopener">https://www.independent.co.uk/life-style/gadgets-and-tech/news/elon-musk-artificial-intelligence-ai-singularity-a9640196.html</a>.</em></li> <li><em>3 Public opinion on AI governance | Artificial Intelligence: American Attitudes and Trends. Accessed January 28, 2022. <a href="https://governanceai.github.io/US-Public-Opinion-Report-Jan-2019/public-opinion-on-ai-governance.html" target="_blank" rel="noopener">https://governanceai.github.io/US-Public-Opinion-Report-Jan-2019/public-opinion-on-ai-governance.html</a>.</em></li> <li><em>Lacker, Kevin. 2020. “Giving GPT-3 a Turing Test.” Kevin Lacker’s blog. <a href="https://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html" target="_blank" rel="noopener">https://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html</a>.</em></li></ul> <p><em>The original post can be found here: <a href="https://www.opinosis-analytics.com/blog/ai-job-loss/" target="_blank" rel="noopener">Will AI Kill Your Job?</a></em></p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/will-ai-kill-your-job/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">6806</post-id> </item> <item> <title>A Gentle Introduction to Deep Neural Networks with Python</title> <link>https://kavita-ganesan.com/neural-network-intro/</link> <comments>https://kavita-ganesan.com/neural-network-intro/#respond</comments> <dc:creator><![CDATA[Kavita Ganesan]]></dc:creator> <pubDate>Thu, 27 Jan 2022 21:44:08 +0000</pubDate> <category><![CDATA[AI Implementation]]></category> <guid isPermaLink="false">https://kavita-ganesan.com/?p=6748</guid> <description><![CDATA[This article examines the parts that make up neural networks and deep neural networks, as well as the fundamental different types of models (e.g. regression), their constituent parts (and how they contribute to model accuracy), and which tasks they are designed to learn.]]></description> <content:encoded><![CDATA[<p><em>This is a guest post from Andrew Ferlitsch, author of <a href="https://www.manning.com/books/deep-learning-patterns-and-practices">Deep Learning Patterns and Practices</a>. It provides an introduction to deep neural networks in Python. Andrew is an expert on computer vision, deep learning, and operationalizing ML in production at Google Cloud AI Developer Relations.</em></p> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-817x1024.png" alt="" class="wp-image-6795" width="174" height="218" srcset="https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-817x1024.png 817w, https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-239x300.png 239w, https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-120x150.png 120w, https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-768x963.png 768w, https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-1225x1536.png 1225w, https://kavita-ganesan.com/wp-content/uploads/Ferlitsch-DL-HI-1633x2048.png 1633w" sizes="(max-width: 174px) 100vw, 174px" /></figure></div> <p class="has-black-color has-text-color has-background" style="background-color:#eeeeee">This article examines the parts that make up neural networks and deep neural networks, as well as the fundamental different types of models (e.g. regression), their constituent parts (and how they contribute to model accuracy), and which tasks they are designed to learn. This article is meant for machine learning engineers who are familiar with Python and deep learning and want to get a thorough intro to the parts and functions of deep neural networks and related models.</p> <div class="wp-block-uagb-table-of-contents uagb-toc__align-left uagb-toc__columns-1 uagb-block-fc090491 " data-scroll= "1" data-offset= "30" style="" > <div class="uagb-toc__wrap"> <div class="uagb-toc__title"> Table Of Contents </div> <div class="uagb-toc__list-wrap"> <ol class="uagb-toc__list"><li class="uagb-toc__list"><a href="#introduction-to-neural-networks-in-python" class="uagb-toc-link__trigger">Introduction to Neural Networks in Python</a><li class="uagb-toc__list"><a href="#neural-network-basics" class="uagb-toc-link__trigger">Neural Network Basics</a><ul class="uagb-toc__list"><li class="uagb-toc__list"><a href="#input-layer" class="uagb-toc-link__trigger">Input Layer</a></li></ul></li><li class="uagb-toc__list"><a href="#deep-neural-networks-dnn" class="uagb-toc-link__trigger">Deep Neural Networks (DNN)</a><ul class="uagb-toc__list"><li class="uagb-toc__list"><a href="#feed-forward-networks" class="uagb-toc-link__trigger">Feed Forward networks</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#sequential-api-method" class="uagb-toc-link__trigger">Sequential API Method</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#functional-api-method" class="uagb-toc-link__trigger">Functional API Method</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#input-shape-vs-input-layer" class="uagb-toc-link__trigger">Input Shape vs Input Layer</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#dense-layer" class="uagb-toc-link__trigger">Dense Layer</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#activation-functions" class="uagb-toc-link__trigger">Activation Functions</a><li class="uagb-toc__list"><li class="uagb-toc__list"><a href="#shorthand-syntax" class="uagb-toc-link__trigger">Shorthand Syntax</a></ul></ul></ol> </div> </div> </div> <p></p> <h2 class="wp-block-heading"><strong>Introduction to Neural Networks in Python</strong></h2> <p>We will start this article with some basics on neural networks. First, we will cover the input layer to a neural network, then how this is connected to an output layer, and then how hidden layers are added in-between to become what is called a deep neural network. From there, we cover how the layers are made of nodes, how these nodes learn, and how layers are connected to each other to form fully connected neural networks.</p> <p>We will also cover the fundamental different types of models. That is, there are different model types, such as regression and classification, which learn different types of tasks. Depending on the task you want to learn, determines the model type you will design a model for.</p> <p>We will also cover the fundamentals of weights, biases, activations and optimizers, and how they contribute to the accuracy of the model.</p> <h2 class="wp-block-heading"><strong>Neural Network Basics</strong></h2> <p>We will start with some basics on neural networks. First, we will cover the input layer to a neural network, then how this is connected to an output layer, and then how hidden layers are added in-between to become what is called a deep neural network. From there, we cover how the layers are made of nodes, what nodes do, and how layers are connected to each other to form fully connected neural networks.</p> <h3 class="wp-block-heading"><strong>Input Layer</strong></h3> <p>The input layer to a neural network takes numbers! All the input data is converted to numbers. Everything is a number. The text becomes numbers, speech becomes numbers, pictures become numbers, and things that are already numbers are just numbers.</p> <p>Neural networks take numbers either as vectors, matrices, or tensors. These are simply names for the number of dimensions in an array. A vector is a one-dimensional array, such as a list of numbers. A matrix is a two- dimensional array, like the pixels in a black and white image. And a tensor is any array of three or more dimensions. For example, a three dimensional array is a stack of matrices where each matrix is the same dimension. That’s it.</p> <div class="wp-block-image"><figure class="aligncenter size-full is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch.png" alt="A Gentle Introduction to Deep Neural Networks with Python" class="wp-image-6794" width="437" height="472" srcset="https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch.png 1746w, https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch-278x300.png 278w, https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch-947x1024.png 947w, https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch-139x150.png 139w, https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch-768x830.png 768w, https://kavita-ganesan.com/wp-content/uploads/CH02_F01_Ferlitsch-1421x1536.png 1421w" sizes="(max-width: 437px) 100vw, 437px" /></figure></div> <p class="has-text-align-center"><strong>Fig. 1 Comparison of array shapes and corresponding names in deep learning.</strong></p> <p>Speaking of numbers, you might have heard terms like normalization or standardization. In standardization the numbers are converted to be centered around a mean of zero, with one standard deviation on each side of the mean. If you’re saying, ‘I don’t do statistics’ right about now, I know how you feel. But don’t worry. Packages like scikit-learn and numpy have library calls that do this for you. Standardization is basically a button to push, and it doesn’t even need a lever, so there are no parameters to set.</p> <p>Speaking of packages, you’re going to be using a lot of numpy. What is numpy and why is it so popular? Given the interpretive nature of Python, the language handles large arrays poorly. Like really big, super big arrays of numbers – thousands, tens of thousands, millions of numbers. Think of Carl Sagan’s infamous quote on the size of the Universe – “billions and billions of stars.” That’s a tensor!</p> <p>One day a C programmer got the idea to write, in low-level C, a high performance implementation for handling super big arrays, and then added an external Python wrapper. Numpy was born. Today numpy is a class with lots of useful methods and properties, like the property shape which tells you the shape (or dimensions) of the array, and the where() method which allows you to do SQL-like queries on your super big array.</p> <p>All Python machine learning frameworks, including TensorFlow and PyTorch, will take as input on the input layer a numpy multidimensional array. And speaking of C, or Java, or C+, …, the input layer in a neural network is just like the parameters passed to a function in a programming language. That’s it.</p> <p>Let’s get started by installing Python packages you will need. I assume you have <a href="https://www.python.org/downloads/">Python installed</a> (version 3.X). Whether you directly installed it, or it got installed as part of a larger package, like <a href="https://www.anaconda.com/what-is-anaconda/">Anaconda</a>, you got with it a nifty command-like tool called pip. This tool is used to install any Python package you will ever need again, from a single command invocation. You use pip install and then the name of the package. It goes to the global repository PyPi of Python packages and downloads and installs the package for you. It’s quite easy.</p> <p>We want to start off by downloading and installing the Tensorflow framework, and the numpy package. Guess what their names are in the registry, tensorflow and numpy – thankfully very obvious. Let’s do it together. Go to the command line and issue the following:</p> <pre class="wp-block-code"><code lang="python" class="language-python">cmd> pip install tensorflowcmd> pip install numpy</code></pre> <p>With Tensorflow 2.0, Keras is built-in and the recommended model API, referred to now as TF.Keras.</p> <p>TF.Keras is based on object oriented programming with a collection of classes and associated methods and properties. Let’s start simply. Say we have a dataset of housing data. Each row has fourteen columns of data. One column has the sale price of a home. We are going to call that the “label”. The other thirteen columns have information about the house, such as the square footage and property tax. It’s all numbers. We are going to call those the “features”. What we want to do is “learn” to predict (or estimate) the “label” from the “features”. Now before we had all this compute power and these awesome machine learning frameworks, data analysts did this stuff by hand or by using formulas in an Excel spreadsheet with some amount of data and lots and lots of linear algebra.We, however, will use Keras and TensorFlow.</p> <p>We will start by first importing the Keras module from TensorFlow, and then instantiate an Input class object. For this class object, we define the shape or dimensions of the input. In our example, the input is a one-dimensional array (a vector) of 13 elements, one for each feature.</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import InputInput(shape=(13,))</code></pre> <p>When you run the above two lines in a notebook, you will see the output:</p> <pre class="wp-block-preformatted"><<strong>tf.Tensor</strong> 'input_1:0' shape=(?, 13) dtype=float32></pre> <p>This is showing you what Input(shape=(13,)) evaluates to. It produces a tensor object by the name ‘input_1:0’. This name will be useful later in assisting you in debugging your models. The ‘?’ in shape shows that the input object takes an unbounded number of entries (your examples or rows) of 13 elements each. That is, at run-time it will bind the number of one-dimensional vectors of 13 elements to the actual number of examples (rows) you pass in, referred to as the (mini) batch size. The ‘dtype’ shows the default data type of the elements, which in this case is a 32-bit float (single precision).</p> <p class="has-background" style="background-color:#eeeeee">Take 40% off <a href="https://www.manning.com/books/deep-learning-patterns-and-practices"><em>Deep Learning Patterns and Practices</em></a> by entering <strong>fccferlitsch</strong> into the discount code box at checkout at <a href="https://www.manning.com/">manning.com</a>.</p> <h2 class="wp-block-heading"><strong>Deep Neural Networks (DNN)</strong></h2> <p>DeepMind, Deep Learning, Deep, Deep, Deep. Oh my, what’s all this? Deep in this context just means that the neural network has one or more layers between the input layer and the output layer. Visualize a directed graph in layers of depth. The root nodes are the input layer and the terminal nodes are the output layer. The layers in between are known as the hidden or deep layers. So a four-layer DNN architecture would look like this:</p> <pre class="wp-block-preformatted">input layerhidden layerhidden layeroutput layer</pre> <p>To get started, we’ll assume every neural network node in every layer, except the output layer, is the same type of neural network node. And that every node on each layer is connected to every other node on the next layer. This is known as a fully connected neural network (FCNN), as depicted in figure 2. For example, if the input layer has three nodes and the next (hidden) layer has four nodes, then each node on the first layer is connected to all four nodes on the next layer for a total of 12 (3×4) connections.</p> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch-1024x706.png" alt="A Gentle Introduction to Deep Neural Networks with Python" class="wp-image-6793" width="512" height="353" srcset="https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch-1024x706.png 1024w, https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch-300x207.png 300w, https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch-150x103.png 150w, https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch-768x529.png 768w, https://kavita-ganesan.com/wp-content/uploads/CH02_F02_Ferlitsch.png 1336w" sizes="(max-width: 512px) 100vw, 512px" /></figure></div> <p class="has-text-align-center"><strong>Fig. 2 Deep neural networks have one or more hidden layers between the input and output layers. This is a fully-connected network, so the nodes at each level are all connected to each other.</strong></p> <h3 class="wp-block-heading"><strong>Feed Forward networks</strong></h3> <p>The DNN and Convolutional Neural Network (CNN), are known as feed forward neural networks. Feed forward means that data moves through the network sequentially, in one direction, from input to output layer). This is analogous to a function in procedural programming. The inputs are passed as parameters in the input layer, the function performs a sequenced set of actions based on the inputs (in the hidden layers) and outputs a result (the output layer).</p> <p>When coding a forward feed network in TF.Keras, you will see two distinctive styles in blogs and other tutorials. I will briefly touch on both so when you see a code snippet in one style you can translate it to the other.</p> <h3 class="wp-block-heading"><strong>Sequential API Method</strong></h3> <p>The Sequential API method is easier to read and follow for beginners, but the trade-off is that it is less flexible. Essentially, you create an empty forward feed neural network with the Sequential class object, and then “add” one layer at a time, until the output layer. In the examples below, the ellipses represent pseudo code.</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import Sequentialmodel = Sequential()model.add( ...the first layer... )model.add( ...the next layer... )model.add( ...the output layer... )</code></pre> <p class="has-background" style="background-color:#eeeeee"><strong>A Create an empty model.</strong><br><strong>B Placeholders for adding layers in sequential order.</strong></p> <p>Alternatively, the layers can be specified in sequential order as a list passed as a parameter when instantiating the Sequential class object.</p> <pre class="wp-block-code"><code lang="python" class="language-python">model = Sequential([ ...the first layer..., ...the next layer..., ...the output layer... ])</code></pre> <p>So, you might ask, when would one use the add() method versus specifying as a list in the instantiation of the Sequential object. Well, both methods generate the same model and behavior, so it’s a matter of personal preference. For myself, I tend to use the more verbose add() method in instructional and demonstration material for clarity. But, if I am writing code for production, I will use the sparser list method, where I can visualize and edit the code more easily.</p> <h3 class="wp-block-heading"><strong>Functional API Method</strong></h3> <p>The Functional API method is more advanced, allowing you to construct models that are non-sequential in flow –such as branches, skip links, and multiple inputs and outputs. You build the layers separately and then “tie” them together. This latter step gives you the freedom to connect layers in creative ways. Essentially, for a forward feed neural network, you create the layers, bind them to another layer(s), and then pull all the layers together in a final instantiation of a Model class object.</p> <pre class="wp-block-code"><code lang="python" class="language-python">input = layers.(...the first layer...)hidden = layers.(...the next layer...)( ...the layer to bind to... )output = layers.(...the output layer...)( /the layer to bind to... )model = Model(input, output)</code></pre> <p></p> <h3 class="wp-block-heading"><strong>Input Shape vs Input Layer</strong></h3> <p>The input shape and input layer can be confusing at first. They are not the same thing. More specifically, the number of nodes in the input layer does not need to match the shape of the input vector. That’s because every element in the input vector will be passed to every node in the input layer, as depicted in figure 2a.</p> <div class="wp-block-image"><figure class="aligncenter size-full is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/CH02_F03_Ferlitsch.png" alt="A Gentle Introduction to Deep Neural Networks with Python" class="wp-image-6792" width="390" height="372" srcset="https://kavita-ganesan.com/wp-content/uploads/CH02_F03_Ferlitsch.png 780w, https://kavita-ganesan.com/wp-content/uploads/CH02_F03_Ferlitsch-300x286.png 300w, https://kavita-ganesan.com/wp-content/uploads/CH02_F03_Ferlitsch-150x143.png 150w, https://kavita-ganesan.com/wp-content/uploads/CH02_F03_Ferlitsch-768x732.png 768w" sizes="(max-width: 390px) 100vw, 390px" /></figure></div> <p class="has-text-align-center"><strong>Fig. 2a Shows the difference between the input (shape) and input layer and how every element in the input is connected to every node in the input layer.</strong></p> <p>For example, if our input layer is ten nodes, and we use our earlier example of a thirteen-element input vector, we will have 130 connections (10 x 13) between the input vector and the input layer.</p> <p>Each one of these connections between an element in the input vector and a node in the input layer will have a weight and each node in the input layer has a bias. Think of each connection between the input vector and input layer, as well as connections between layers, as sending a signal forward in how strongly it believes the input value will contribute to what the model predictions. We need to have a measurement of the strength of this signal, and that is what the weight does. It is a coefficient that is multiplied against the input value for the input layer, and previous value for subsequent layers. Now each one of these connections is like a vector on an x-y plane. Ideally, we would want each of these vectors to cross the y-axis at the same central point, e.g., 0 origin. But they don’t. To make the vectors relative to each other, the bias is the offset of each vector from the central point on the y-axis.</p> <p>The weights and biases are what the neural network will “learn” during training. The weights and biases are also referred to as parameters. That is, these values stay with the model after it is trained. This operation will otherwise be invisible to you.</p> <h3 class="wp-block-heading"><strong>Dense Layer</strong></h3> <p>In TF.Keras, layers in a fully connected neural network (FCNN) are called Dense layers. A Dense layer is defined as having an “n” number of nodes, and is fully connected to the previous layer. Let’s continue and define in TF.Keras a three layer neural network, using the Sequential API method, for our example. Our input layer will be ten nodes, and take as input a thirteen element vector (i.e., the thirteen features), which will be connected to a second (hidden) layer of ten nodes, which will then be connected to a third (output) layer of one node. Our output layer only needs to be one node, since it will be outputting a single real value (e.g. – the predicted price of the house). This is an example where we are going to use a neural network as a regressor. That means, the neural network will output a single real number.</p> <pre class="wp-block-preformatted">input layer = 10 nodeshidden layer = 10 nodesoutput layer = 1 node</pre> <p>For input and hidden layers, we can pick any number of nodes. The more nodes we have, the better the neural network can learn, but more nodes means more complexity and more time in training and predicting.</p> <p>In the following code example, we have three add() calls to the class object Dense(). The add() method “adds” the layers in the same sequential order we specified them in. The first (positional) parameter is the number of nodes, ten in the first and second layer and one in the third layer. Notice how in the first Dense() layer we added the (keyword) parameter input_shape. This is where we will define the input vector and connect it to the first (input) layer in a single instantiation of Dense().</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import Sequentialfrom tensorflow.keras.layers import Densemodel = Sequential()# Add the first (input) layer (10 nodes) with input shape 13 element vector (1D).model.add(Dense(10, input_shape=(13,)))# Add the second (hidden) layer of 10 nodes.model.add(Dense(10))# Add the third (output) layer of 1 node.model.add(Dense(1))</code></pre> <p>Alternatively, we can define the sequential sequence of the layers as a list parameter when instantiating the Sequential class object.</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import Sequentialfrom tensorflow.keras.layers import Densemodel = Sequential([ # Add the first (input) layer (10 nodes) Dense(10, input_shape=(13,)), # Add the second (hidden) layer of 10 nodes. Dense(10), # Add the third (output) layer of 1 node. Dense(1) ])</code></pre> <p>Let’s now do the same but use the Functional API method. We start by creating an input vector by instantiating an Input class object. The (positional) parameter to the Input() object is the shape of the input, which can be a vector, matrix or tensor. In our example, we have a vector that is thirteen elements long. So our shape is (13,). I am sure you noticed the trailing comma. That’s to overcome a quirk in Python. Without the comma, a (13) is evaluated as an expression. That is, the integer value 13 is surrounded by a parenthesis. Adding a comma will tell the interpreter this is a tuple (an ordered set of values).</p> <p>Next, we create the input layer by instantiating a Dense class object. The positional parameter to the Dense() object is the number of nodes; which in our example is ten. Note the peculiar syntax that follows with a (inputs). The Dense() object is a callable. That is, the object returned by instantiating the Dense() object can be callable as a function. So we call it as a function, and in this case, the function takes as a (positional) parameter the input vector (or layer output) to connect it to; hence we pass it inputs so the input vector is bound to the ten node input layer.</p> <p>Next, we create the hidden layer by instantiating another Dense() object with ten nodes, and using it as a callable, we (fully) connect it to the input layer.</p> <p>Then we create the output layer by instantiating another Dense() object with one node, and using it as a callable, we (fully) connect it to the hidden layer.</p> <p>Finally, we put it altogether by instantiating a Model class object, passing it the (positional) parameters for the input vector and output layer. Remember, all the other layers in-between are already connected so we don’t need to specify them when instantiating the Model() object.</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import Input, Modelfrom tensorflow.keras.layers import Dense inputs = Input((13,))input = Dense(10)(inputs)hidden = Dense(10)(input)output = Dense(1)(hidden)model = Model(inputs, output)</code></pre> <h3 class="wp-block-heading"><strong>Activation Functions</strong></h3> <p>When training or predicting (inference), each node in a layer will output a value to the nodes in the next layer. We don’t always want to pass the value ‘as-is’, but instead sometimes we want to change the value in some manner. This process is called an activation function. Think of a function that returns some result, like return result. In the case of an activation function, instead of returning result, we would return the result of passing the result value to another (activation) function, like return A(result), where A() is the activation function. Conceptually, you can think of this as:</p> <pre class="wp-block-code"><code lang="python" class="language-python">def layer(params): """ inside are the nodes """ result = some_calculations return A(result) def A(result): """ modifies the result """ return some_modified_value_of_result</code></pre> <p>Activation functions assist neural networks in learning faster and better. By default, when no activation function is specified, the values from one layer are passed as-is (unchanged) to the next layer. The most basic activation function is a step function. If the value is greater than 0, then a 1 is outputted; otherwise a zero. The step function hasn’t been used in a long, long time.</p> <p>Let’s pause for a moment and discuss the purpose of an activation function. You likely have heard the phrase non-linearity. What is this? To me, more importantly, is what it is not?</p> <p>In traditional statistics, we worked in low dimensional space where there was a strong linear correlation between the input space and output space. This correlation could be computed as a polynomial transformation of the input that, when transformed, had a linear correlation to the output. The most fundamental example is the slope of a line, which is represented as y = mx + b. In this case, x and y are coordinates of the line, and we want to fit the value of m, the slope, and b, where the line intercepts the y access.</p> <p>In deep learning, we work in high dimensional space where there is substantial non-linearity between the input space and output space. What is non-linearity? It means that an input is not (near) uniformly related to an output based on a polynomial transformation of the input. For example, let’s say one’s property tax is a fixed percentage rate (r) of the house value. In this case, the property tax can be represented by a function that multiplies the rate by the house value — thus having a linear (i.e., straight line) relationship between value (input) and property tax (output).</p> <pre class="wp-block-preformatted">tax = F(<strong>value</strong>) = r * <strong>value</strong></pre> <p>Let’s look at the logarithmic scale for measuring earthquakes, where an increase of one, means the power released is ten times greater. For example, an earthquake of 4 is 10 times stronger than a 3. By applying a logarithmic transform to the input power we have a linear relationship between power and scale.</p> <pre class="wp-block-preformatted">scale = F(power) = log(power)</pre> <p>In a non-linear relationship, sequences within the input have different linear relationships to the output, and in deep learning we want to learn both the separation points as well as the linear functions for each input sequence. For example, consider age vs. income to demonstrate a non-linear relationship. In general, toddlers have no income, grade-school children have an allowance, early-teens earn an allowance + money for chores, later teens earn money from jobs, and then when they go to college their income drops to zero! After college, their income gradually increases until retirement, when it becomes fixed. We could model this nonlinearity as sequences across age and learn a linear function for each sequence, such as depicted below.</p> <pre class="wp-block-preformatted">income = F1(age) = 0 for age [0..5]income = F2(age) = c1 for age[6..9]income = F3(age) = c1 + (w1 * age) for age[10..15]income = F4(age) = (w2 * age) for age[16..18]income = F5(age) = 0 for age[19..22]income = F6(age) = (w3 * age) for age[23..64]income = F7(age) = c2 for age [65+]</pre> <p>Activation functions assist in finding the non-linear separations and corresponding clustering of nodes within input sequences which then learn the (near) linear relationship to the output.</p> <p>There are three activation functions you will use most of the time: the rectified linear unit (ReLU); sigmoid; softmax. We will start with the ReLU, since it is the one that is most used in all but the output layer of a model. The sigmoid and softmax activation we will then cover when we look at how different model types affect the design of the output layer.</p> <p>The rectified linear unit, as depicted in figure 3, passes values greater than zero as-is (unchanged); otherwise zero (no signal).</p> <div class="wp-block-image"><figure class="aligncenter size-large is-resized"><img loading="lazy" decoding="async" src="https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-1024x847.png" alt="A Gentle Introduction to Deep Neural Networks with Python" class="wp-image-6791" width="512" height="424" srcset="https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-1024x847.png 1024w, https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-300x248.png 300w, https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-150x124.png 150w, https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-768x635.png 768w, https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-1536x1271.png 1536w, https://kavita-ganesan.com/wp-content/uploads/CH02_F04_Ferlitsch-2048x1694.png 2048w" sizes="(max-width: 512px) 100vw, 512px" /><figcaption><strong>Fig. 3 The function for a rectified linear unit clips all negative values to zero. In essence, any negative value is the same as no signal ~ zero.</strong></figcaption></figure></div> <p class="has-text-align-center"></p> <p>The rectified linear unit is generally used between layers. While early researchers used different activation functions, such as a hyperbolic tangent, between layers, researchers found that the ReLU produced the best result in training a model. In our example, we will add a rectified linear unit between each layer.</p> <pre class="wp-block-code"><code lang="python" class="language-python"> model = Sequential()# Add the first (input) layer (10 nodes) with input shape 13 element vector (1D).model.add(Dense(10, input_shape=(13,)))# Pass the output from the input layer through a rectified linear unit activation # function.model.add(ReLU())# Add the second (hidden) layer (10 nodes).model.add(Dense(10))# Pass the output from the input layer through a rectified linear unit activation # function.model.add(ReLU())# Add the third (output) layer of 1 node.model.add(Dense(1))</code></pre> <p>Let’s take a look inside our model object and see if we constructed what we think we did. You can do this using the summary() method. It will show in sequential order a summary of each layer.</p> <pre class="wp-block-code"><code lang="python" class="language-python">model.summary()</code></pre> <pre class="wp-block-preformatted">Layer (type) Output Shape Param # =================================================================dense_56 (Dense) (None, 10) 140 _________________________________________________________________re_lu_18 (ReLU) (None, 10) 0 _________________________________________________________________dense_57 (Dense) (None, 10) 110 _________________________________________________________________re_lu_19 (ReLU) (None, 10) 0 _________________________________________________________________dense_58 (Dense) (None, 1) 11 =================================================================Total params: 261Trainable params: 261Non-trainable params: 0_________________________________________________________________</pre> <p>For this code example, you see the summary starts with a Dense layer of ten nodes (input layer), followed by a ReLU activation function, followed by a second Dense layer (hidden) of ten nodes, followed by a ReLU activation function, and finally followed by a Dense layer (output) of one node. So, yes, we got what we expected.</p> <p>Next, let’s look at the parameter field in the summary. See how, for the input layer, it shows 140 parameters. How is that calculated? We have 13 inputs and 10 nodes, so 13 x 10 is 130. Where does 140 come from? Each connection between the inputs and each node has a weight, which adds up to 130. But each node has an additional bias. That’s ten nodes, so 130 + 10 = 140. As I’ve said, it’s the weights and biases that the neural network will “learn” during training. A bias is a learned offset, conceptually equivalent to the y-intercept (b) in the slope of a line, which is where the line intercepts the y-axis:</p> <pre class="wp-block-preformatted">y = b + mx</pre> <p>At the next (hidden) layer you see 110 params. That’s ten outputs from the input layer connected to each of the ten nodes from the hidden layer (10×10) plus the ten biases for the nodes in the hidden layers, for a total of 110 parameters to “learn”.</p> <h3 class="wp-block-heading"><strong>Shorthand Syntax</strong></h3> <p>TF.Keras provides a shorthand syntax when specifying layers. You don’t actually need to separately specify activation functions between layers, as we did above. Instead, you can specify the activation function as a (keyword) parameter when instantiating a Dense() layer.</p> <p>You might ask, why not then simply always use the shorthand syntax? As you will see later in the book, where in today’s model architecture the activation function is preceded by another intermediate layer — batch normalization, or precedes the layer altogether — pre-activation batch normalization.</p> <p>The code example below does exactly the same as the code above.</p> <pre class="wp-block-code"><code lang="python" class="language-python">from tensorflow.keras import Sequentialfrom tensorflow.keras.layers import Dense model = Sequential()# Add the first (input) layer (10 nodes) with input shape 13 element vector (1D).model.add(Dense(10, input_shape=(13,), activation='relu'))# Add the second (hidden) layer (10 nodes).model.add(Dense(10, activation='relu'))# Add the third (output) layer of 1 node.model.add(Dense(1))#A The activation function is specified as a keyword parameter in the layer.</code></pre> <p>Let’s call the summary() method on this model.</p> <pre class="wp-block-code"><code class="">model.summary()</code></pre> <pre class="wp-block-preformatted">Layer (type) Output Shape Param # =================================================================dense_56 (Dense) (None, 10) 140 _________________________________________________________________re_lu_18 (ReLU) (None, 10) 0 _________________________________________________________________dense_57 (Dense) (None, 10) 110 _________________________________________________________________re_lu_19 (ReLU) (None, 10) 0 _________________________________________________________________dense_58 (Dense) (None, 1) 11 ================================================================= Total params: 261Trainable params: 261Non-trainable params: 0</pre> <p>Hum, you don’t see the activations between the layers as you did in the earlier example. Why not? It’s a quirk in how the summary() method displays output. They are still there.</p> <p><strong>Improving accuracy with optimizer</strong></p> <p>Once you’ve completed building the forward feed portion of your neural network, as we have for our simple example, we now need to add a few things for training the model. This is done with the compile() method. This step adds the backward propagation during training. Let’s define and explore this concept.</p> <p>Each time we send data (or a batch of data) forward through the neural network, the neural network calculates the errors in the predicted results (known as the loss) from the actual values (called labels) and uses that information to incrementally adjust the weights and biases of the nodes. This, for a model, is the process of learning.</p> <p>The calculation of the error, as I’ve said, is called a loss. It can be calculated in many different ways. Since we designed our example neural network to be a regresser (meaning that the output, house price, is a real value), we want to use a loss function that is best suited for a regresser. Generally, for this type of neural network, we use the Mean Square Error method of calculating a loss. In Keras, the compile() method takes a (keyword) parameter loss where we can specify how we want to calculate the loss. We are going to pass it the value ‘mse’ for Mean Square Error.</p> <p>The next step in the process is the optimizer that occurs during backward propagation. The optimizer is based on gradient descent; where different variations of the gradient descent algorithm can be selected. These terms can be hard to understand at first. Essentially, each time we pass data through the neural network we use the calculated loss to decide how much to change the weights and biases in the layers by. The goal is to gradually get closer and closer to the correct values for the weights and biases to accurately predict or estimate the “label” for each example. This process of progressively getting closer and closer to the accurate values is called convergence. The job of the optimizer is to calculate the updates to the weights to progressively get closer to the accurate values to reach convergence.</p> <p>As the loss gradually decreases we are converging and once the loss plateaus out, we have convergence, and the result is the accuracy of the neural network. Before using gradient descent, the methods used by early AI researchers could take years on a supercomputer to find convergence on a non-trivial problem. After the discovery of using the gradient descent algorithm, this time reduced to days, hours and even just minutes on ordinary compute power. Let’s skip the math and just say that gradient descent is the data scientist’s pixie dust that makes convergence possible.</p> <p>For our regressor neural network, we will use the <code>rmsprop</code> method (root mean square property).</p> <pre class="wp-block-code"><code class="">model.compile(loss='mse', optimizer='rmsprop')</code></pre> <p>Now we have completed building your first ‘trainable’ neural network.</p> <p>That’s all for now. If you want to learn more about the book, check it out on Manning’s liveBook platform <a href="https://livebook.manning.com/book/deep-learning-patterns-and-practices?origin=product-look-inside">here</a>.</p>]]></content:encoded> <wfw:commentRss>https://kavita-ganesan.com/neural-network-intro/feed/</wfw:commentRss> <slash:comments>0</slash:comments> <post-id xmlns="com-wordpress:feed-additions:1">6748</post-id> </item> </channel></rss> 
