<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Joseph Cha on Medium]]></title>
        <description><![CDATA[Stories by Joseph Cha on Medium]]></description>
        <link>https://medium.com/@joseph_cha?source=rss-31a888191d05------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*ARIEsCbfTzm5y6WRHVzUAw.jpeg</url>
            <title>Stories by Joseph Cha on Medium</title>
            <link>https://medium.com/@joseph_cha?source=rss-31a888191d05------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 22:25:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@joseph_cha/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[On “AI Pilot Purgatory”]]></title>
            <link>https://medium.datadriveninvestor.com/on-ai-pilot-purgatory-81787237294f?source=rss-31a888191d05------2</link>
            <guid isPermaLink="false">https://medium.com/p/81787237294f</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Joseph Cha]]></dc:creator>
            <pubDate>Thu, 09 Jan 2020 20:06:01 GMT</pubDate>
            <atom:updated>2020-01-09T20:06:01.354Z</atom:updated>
            <content:encoded><![CDATA[<p>(And how to avoid it)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6bWO30wqxb_FwFdwX3zVHQ.jpeg" /></figure><h3>## TL; DR</h3><p>“How to Avoid AI Pilot Purgatory Cheat Sheet”</p><p>Following best practices → increased likelihood of production of useful tools</p><ul><li>PoC and MVP approaches prevent use-case disconnect</li><li><strong>Solid Productionisation</strong> (setting access rights, logging the users’ activity, data governance, maintaining the code base, scalability, releasing updates and setting a common environment for the developers, etc.<strong>) is a key differentiator</strong></li><li>Maintenance of the model is equally as challenging as researching the model.</li><li>Preparing responses for common problems, such as concept drift, model bias, resistance to adoption by teams.</li><li>Adding computational thinking to the business organization → improved use case discovery and increased adoption of the solution</li><li>AI/ML projects are risky, some will fail, but the upside to successful AI implementation is real</li></ul><h3>## On AI Pilot Purgatory (and How to Avoid It)</h3><p>I read a McKinsey<a href="https://www.mckinsey.com/featured-insights/artificial-intelligence/ai-adoption-advances-but-foundational-barriers-remain"> study</a> entitled “AI adoption advances, but foundational barriers remain” by Michael Chui and Sankalp Malhotra. One paragraph stood out:</p><p><em>While most companies have already deployed AI to some extent, few have embedded it into standard operating processes in multiple business units or functions, and about one-third are </em><strong><em>only piloting</em></strong><em> the use of AI. While AI is still in its early days, getting stuck in “pilot purgatory” is a real risk.</em></p><p>I imagine it must be extremely frustrating to a company or other organization that invested in building and deploying an AI model only to watch it languish in “AI pilot purgatory,” where at best a handful of people use it, and the dreams of scale are not realized. There is a great deal of buzzword-filled hype about artificial intelligence in the media. The fact of the matter is that if it’s properly built, deployed and maintained, AI does offer considerable value and competitive advantage. But if an AI model is not properly built, deployed, and maintained, there is a real risk that an organization will end up with a lonely AI sitting in their servers or in the cloud, the result of big dreams and good intentions, helping no one.</p><p><a href="https://www.datadriveninvestor.com/2019/02/19/artificial-intelligence-trends-to-watch-this-year/">5 Artificial Intelligence Trends To Watch This Year | Data Driven Investor</a></p><p>By the way, Merriam-Webster defines “purgatory” as <strong>“</strong>a place or state of temporary suffering or misery.”</p><p>Is the adoption of AI solutions an issue? How can one prevent “AI Pilot Purgatory”? Actually, introducing <strong>any</strong> type of major initiative to a company can be challenging, not just AI projects. What are the similarities between introducing a more common type of major initiative and introducing an AI model to an organization?</p><p>I consulted some local experts about this issue: Filip Stachura and Marcin Dubel (in the following post).</p><h3>Filip Stachura</h3><iframe src="https://www.instagram.com/p/B7BhmZzlLeb/embed/?cr=1&amp;amp;rd=https%3A%2F%2Fwww.instagram.com" width="658" height="882" frameborder="0" scrolling="no"><a href="https://medium.com/media/d6067b06c2f6ffe4ee6c9ffc427524cc/href">https://medium.com/media/d6067b06c2f6ffe4ee6c9ffc427524cc/href</a></iframe><ol><li><strong>Most AI/ML projects are risky. </strong>To some extent this is normal that some of the hypotheses are going to fail (e.g. technology is too expensive, or results are not satisfying for business)</li><li>Even if you get good business results, <strong>productionization/operationalization of models is hard</strong>! You need to plug models into your data environment and make sure your model won’t do anything stupid in production <strong>and </strong>prepare for a moment it will. Here is an <a href="https://www.forbes.com/sites/mzhang/2015/07/01/google-photos-tags-two-african-americans-as-gorillas-through-facial-recognition-software/#3d35d997713d">example</a>.</li><li>You need different skills in the team to create the model (research) and to take care of it (data engineers / IT). Maintenance is as hard as research, especially if you have events like data drift.</li><li>Data going into the model needs to be clean. This is a big problem. Data scientists / Researchers clean their data, but very often this is a one-time event. You need proper <strong>data governance</strong> to make data clean for an AI model in production.</li><li>Current employees can be afraid of losing their jobs and hamper adoption</li><li>Current employees can be truly less relevant — how are you going to help them to adapt?</li><li>The world can change and one day the model can start to behave badly (This is called <a href="https://en.wikipedia.org/wiki/Concept_drift"><strong>concept drift</strong></a>. Such a situation is common but easily mitigated with planning).</li><li>Who is going to validate the results of the model during initial tests and later on?</li><li>How is the model going to be validated?</li></ol><h3>Discussion</h3><p>Interesting to note that it seems that many of the challenges listed above are more human than technological in nature. Professor Mihnea Moldoveanu from the University of Toronto puts it well in “<a href="https://hbr.org/2019/03/why-ai-underperforms-and-what-companies-can-do-about-it">Why AI Underperforms and What Companies Can Do About It</a>”:</p><blockquote>“Why is the gap between companies’ AI ambition and their actual adoption so large? <strong>The answer is not primarily technical.</strong> It is organizational and cultural. A massive skills and language gap has emerged between key organizational decision makers and their AI teams… And it is growing, not shrinking.”</blockquote><p>Filip mentions “productionisation” in point #2 above. We talk about productionisation a lot in our blogposts, such as <a href="https://appsilon.com/ai-transformation-of-insurance/">here</a> and <a href="https://appsilon.com/decision-support-systems-4-how-to-implement-an-ia-solution/">here</a>. Productionisation is basically support and planning by humans to ensure AI model success. Someday there may be AI models that are self-aware and independent, but we are still a long way off from that. Here is a summary of what we mean:</p><p><strong>Data validation</strong>. The initial dataset that you use to train the AI model is important, but this isn’t a one-time event. Typically we receive new data as time goes on, and we want to update the model with the new data to teach it model new things. Automated data verification is therefore necessary, so that we can immediately get alerted if there are problems with the data.</p><p><strong>Interpretability</strong>. Our clients need to be prepared to explain their decisions to the users of their products. Recommendations from a model are only helpful if the decision making process is explainable.</p><p><strong>Reliability and scaling</strong>. When putting a model into production we need a plan for how it will scale. We need to plan for events like spikes in usage or server outage.</p><p><strong>Human augmentation and oversight</strong>. We shouldn’t think of AI as a solution that does everything for you. Instead, think of it as IA — intelligence augmentation, and design how humans can work together with the model and oversee its operation.</p><p><strong>Automated model update</strong>. For successful implementation of an AI model, it needs to learn new things based on new data. This should be fully automated, and not a manual process.</p><p><strong>User interface. </strong>It’s important to match your state-of-the-art AI model with a user interface that the end-users can employ in their tasks without friction. Depending on the use case, this can be an API called by other systems or a<a href="https://appsilon.com/how-we-built-a-shiny-app-for-700-users/"> human-friendly Shiny dashboard</a>.</p><p>Filip also mentions in point #3 above that you need different people to create the AI model and maintain it. It’s a matter of putting the right people in the right place to keep your AI up and running (again, an organizational, rather than technological scenario). Point #4 refers to the practice of strategizing and executing on the collection and validation of data in your organization. For examples of data governance in action, you can read <a href="https://appsilon.com/author/pawel/">Paweł Przytuła</a>’s article “<a href="https://appsilon.com/data-quality/">Data Quality Case Studies: How We Saved Clients Real Money Thanks to Data Validation</a>.”</p><p>There are several challenges that are common enough that you might as well plan for them (bias, rejection of adoption by employees, data problems). Another <a href="https://www.mckinsey.com/featured-insights/artificial-intelligence/global-ai-survey-ai-proves-its-worth-but-few-scale-impact?cid=soc-web">McKinsey survey</a> indicates that “AI Power Users” often acknowledge these risks and challenges, and plan for them.</p><h3>## The future is bright, but challenges remain</h3><p>Implementing AI is risky, complex and challenging, but the potential upside of successful AI adoption is significant. It is certainly worth investing in best practices to ensure success. The same article from McKinsey that alarmed us with the quote about “AI pilot purgatory” also gave us the following quotes:</p><blockquote>Although many AI projects languish in purgatory, a majority of executives whose companies have adopted AI report that it has provided an uptick in revenue in the business areas where it is used, and 44 percent say AI has reduced costs.</blockquote><p>And the growth of AI adoption continues…</p><blockquote>47% of respondents <em>[of their 2019 survey]</em> say their companies have embedded at least one AI capability in their business processes — compared with 20 percent of respondents in a 2017 study who said their companies were using AI in a core part of their business.</blockquote><p>What is your take on “AI Pilot Purgatory”? Please add your thoughts in the comments below.</p><p>Thanks for reading. Follow me on Twitter <a href="https://twitter.com/_joecha_">@_joecha_</a>.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=81787237294f" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/on-ai-pilot-purgatory-81787237294f">On “AI Pilot Purgatory”</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Generate AI and DS Use Cases for Your Business]]></title>
            <link>https://medium.datadriveninvestor.com/how-to-generate-ai-and-ds-use-cases-for-your-business-23835d716f86?source=rss-31a888191d05------2</link>
            <guid isPermaLink="false">https://medium.com/p/23835d716f86</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[mls]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Joseph Cha]]></dc:creator>
            <pubDate>Mon, 18 Nov 2019 18:30:05 GMT</pubDate>
            <atom:updated>2019-11-19T05:08:06.168Z</atom:updated>
            <content:encoded><![CDATA[<p>(If your company name isn’t Google, Facebook, Baidu, etc.)</p><h3>tl; dr</h3><p>Cutting through AI hype to reveal potential value for your organization. How companies can benefit from digital transformation, where and how to look for use cases, and successfully deliver data science projects. Gain a massive competitive advantage and avoid waste by choosing the right tools for the job.</p><h3>Brainstorming Use Cases</h3><p><em>I recently had a conversation with a salesperson from one of the world’s largest electronics manufacturers. He told me that a lot of the factory owners that he chats with are very interested in adding data science, artificial intelligence, and IoT approaches and tools to their operations, but they haven’t a clue where to start. I was a bit surprised! And it’s a fascinating challenge. What are the best ways for a non-software company to generate use cases for digital transformation of their operations? In many cases, a factory or some other organization has been operating successfully using the same methods for quite some time. Change brings disruption. But even an organization that embraces change and disruption may struggle in finding ways to improve their operations with the numerous digital transformation tools available today.</em></p><p><a href="https://www.datadriveninvestor.com/2019/02/07/8-skills-you-need-to-become-a-data-scientist/">8 Skills You Need to Become a Data Scientist | Data Driven Investor</a></p><p><em>How does one </em><strong><em>begin</em></strong><em> such a task? Obviously, a company needs to generate use cases to hand over to an AI tools team to build a Proof of Concept (PoC) model. It’s just </em><strong><em>digital transformation</em></strong><em>. Easy… Right?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/868/1*Et11bztNcO6ILEhbJ8KcGA.jpeg" /></figure><p><em>We asked Marcin Dubel, Warsaw-based Data Scientist extraordinaire and author of the “</em><a href="https://www.r-bloggers.com/super-solutions-for-shiny-architecture-1-of-5-using-session-data/"><em>Super Solutions for Shiny Architecture</em></a><em>” series and “</em><a href="https://www.r-bloggers.com/forget-about-excel-use-these-r-shiny-packages-instead/"><em>Forget about Excel, Use these R Shiny Packages Instead</em></a><em>” articles,</em></p><p><strong>JC: Have you ever helped a client develop use cases for ML, DS, or AI in the form of a brainstorming session? in other words have you ever worked with a firm develop a use case? How do you go about such a task?</strong></p><p>MD: Well it’s an interesting question, Joe. With all of the current hype on powerful machine learning implementations like image recognition, autonomous cars and deep fake videos, many clients are missing the crucial point: digital transformation can be a benefit for virtually all companies and businesses. And this has been happening for years! Extracting value from data or task automation has a long history of successful implementations, that I am proud to be a part of.</p><figure><img alt="SpaceX rocket inside a hangar." src="https://cdn-images-1.medium.com/max/1024/1*iamLmoOnlBmf7tTcf_J_Dw.jpeg" /></figure><p>I see basically <strong>four</strong> use cases that I usually work on with colleagues: (1) <strong>task automation</strong>, (2) <strong>getting value from data</strong>, (3) <strong>machine learning models</strong>, and (4) <strong>decision support systems</strong>. The classification is not strict and usually data science projects are some mix of them. What is also often the case is that we brainstorm with the client the initial idea for improvements in one area, e.g. some automation, and then we spot potential in some data validation or implementing sophisticated analytics. Such opportunities are hard to spot without letting the data science team dig into the data in close collaboration with the client.</p><p>Maybe we can look at some examples of projects for each area, which will make it easier to develop the intuition of how it can be beneficial for a company. Task automation: the popular case here is an employee feeding the data to some Excel file, recalculating the values, generating a report by e.g. copy-pasting the plots to PowerPoint, and sending it to the client as a monthly update. We can transform it into an interactive dashboard automatically pulling the newest data with direct access for the client at any time. And you can cross-out such repetitive tasks from your schedule.</p><p>Getting value from data has two aspects: one can be <strong>monitoring the correctness of data</strong> and the other is looking for the answers <strong>inside</strong>. My colleagues told me about a <a href="https://www.r-bloggers.com/data-quality-case-studies-how-we-saved-clients-real-money-thanks-to-data-validation/">great case of how data validation</a> helped a customer a lot: a new employee misunderstood a task and tried to create a discount fo certain products in the database, but used the wrong +/- sign. So prices would not be reduced but increased! Their data validation tool spotted this action and prevented its implementation.</p><p>Another case is analyzing the data to get some insights. I was once at a crowdfunding conference where some portal creators were presenting the strategies that increase the probability of a project successfully collecting funds. The speakers were aware of the value they have in their databases, but, as they said during the talk: ‘we don’t know how to analyze the data to get proper, evidence-supported answers.’ So, after the talk, I approached them and said that I do know how to do this. The result of the story was a nice report about the drivers of successful crowdfunding projects that they were able to share with their users to increase success rates.</p><p>A typical project also includes such cases as customer segmentation, sentiment analysis of text, detecting trends and time patterns, or analyzing satellite images.</p><figure><img alt="Marcin Dubel discusses a problem with a data scientist and product manager." src="https://cdn-images-1.medium.com/max/1024/1*FQz6PHiPk-MSFOwIjVvjoQ.jpeg" /></figure><p>Recently there has been a boost in machine learning mainly connected with deep neural networks, but also with the development of tools supporting the modelers like using the cloud for heavy computations. That is great, as you know, modern problems require modern solutions. Like switching to renewable energy sources is super important, but raises challenges — solar panels and wind farm energy supply depends heavily on weather conditions. Nevertheless, power plants need to transfer a stable amount of energy to the grid to meet the demand! Combining the weather forecasts with time series machine learning models is a perfect solution in this case that allows for the reduction of energy waste and costs. That can be a game-changer for green energy initiatives.</p><p>And finally, we get to the decision support systems. There are <a href="https://shiny.rstudio.com/">Shiny dashboards</a>, which is a popular framework as it combines web-interface interactivity with the statistical analysis power of the <a href="https://www.r-project.org/about.html">R language</a>. But my colleagues and I have also created tools using <a href="https://plot.ly/dash/">Dash for Python</a> and <a href="https://reactjs.org/">React.js</a> — for each case the proper tools should be used. The point is that such solutions represent the trend of <a href="https://medium.com/authority-magazine/my-message-is-that-ai-and-people-can-be-friends-with-author-alex-bates-chaya-weiner-4a173abde77c">Intelligence Augmentation</a> — providing the decision maker with the support of the newest data, some insights from the model, great visualizations and intuitive interactive interface to trigger the action.</p><p><strong>Andrew Ng recommends gathering two important groups into a room for a use case brainstorm session: the people who know your business, and the people who know AI and Data Science. How does such a meeting work?</strong></p><p>Yes, getting the business perspective into any data project is the crucial part. It is even stated in the definition of the data scientist: the combination of analytical thinking, hacking skills and business perspective. To learn about the exact business needs our experts need to sit down with clients and get into details. We always ask for designating the specific person on the client side who will be available for regular meetings, has enough business knowledge and authority to make decisions. Then it is much easier to communicate and figure out the fruitful outcome of such synergy.</p><p>During such a brainstorm meeting it is an interesting challenge to set the project priorities well. Business people often want to know upfront all of the details and costs of the big projects, like what will be the accuracy of the ML model or how we can connect LinkedIn authentication to the dashboard. It is our job to figure out with the client what is the true, basic need that our solution can satisfy. And after that part, we can work on details and extra features. We use the two well-known concepts for our projects: <strong>Proof of Concept (PoC)</strong> and <strong>Minimal Viable Product (MVP)</strong>. A PoC gives answers like “yes, this problem can be solved using…” or “yes, it can be solved, but we need to gather additional data…” So it is like a quick check, giving the green light for some solutions, making research on what is already on the market. It truly helps for the final solution to be well-designed and thought through. MVP is a piece of quickly delivered solution that is solving the core of the problem but lacks additional features, nevertheless, it is something that will be valuable for the client even if we won’t develop it anymore. Having those concepts in mind helps a lot for business people and data science experts to be on the same page.</p><figure><img alt="A meeting of the minds at a brainstorm session." src="https://cdn-images-1.medium.com/max/1024/1*YRIzc9krg542DE4rH5J_xQ.jpeg" /></figure><p>There is also a third group that is often forgotten but is crucial for all the projects: <strong>the</strong> <strong>users</strong>. No matter if we are talking about some machine learning model, task automation or decision support dashboard there is usually some real person using it, either by triggering the code, reading the output report or clicking in the application. And usually the decision makers are not the ones who will actually use the final product or it might be even built for external clients. We are always interested in who is the end user of our solution and, if possible, ask for a test session and feedback. At the end of the day the user experience is the crucial part of the project success.</p><figure><img alt="Robot astronaut and Human astronaut look at each other." src="https://cdn-images-1.medium.com/max/1024/1*goOnx2Z8vfLHW2Lq5TpENA.jpeg" /></figure><p><strong>Is it true that companies that do not develop AI capability will be “left in the dust” by the competition, so to speak?</strong></p><p>There is some truth to this, but I would not say it so strongly. If your business is delivering good quality and satisfy the clients&#39; needs then there is no need to implement some AI just because its achievements were in the news. Nevertheless, as I’ve mentioned before, data science projects can help the business in many aspects. And usually, there are either some repetitive tasks that are not strictly connected with the core business or some data analysis to be done that can bring new insights. So not using AI or DS can limit the company’s growth and having data science consulting can bring the business to the next level, allowing it to get some competitive advantage.</p><p><strong>McKinsey </strong><a href="https://www.mckinsey.com/featured-insights/artificial-intelligence/ai-adoption-advances-but-foundational-barriers-remain"><strong>released a study</strong></a><strong> that states that most companies never make it out of AI pilot program phase. Is adoption of AI solutions an issue? If so, how can that be remedied?</strong></p><p>I think I see where the problem is. The outcome of the AI or DS projects is often hard to estimate. There is a lot of research included, testing hypotheses, working on improvements. It cannot be stated upfront: “this project will take 65 man-days and the result will be a 13% increase in sales.” No. But this is the beauty of the data science: you can be surprised how it will boost your business. That is why the PoC and MVP concepts are so crucial for the inception of a project. We employ agile and work in close collaboration with the client, so we can adjust quickly and deliver the biggest value first. The point is to organize the project in such a way that something useful and valuable is produced during each week of production. Working the other way round may be the cause of such study results: when after a few months there is still no value delivered by the AI then the project can be decided to be shut down.</p><p>There is also a second challenge with implementing AI solutions. Building a good model is one thing, but productionalization, i.e. making it useful for the business, is a different story. The whole workflow must be set correctly. From feeding and updating data, through the correct, tested, reliable model response (sometimes also the speed of the response is crucial!) to the bug-free user interface. Often the case is in integrating many of the already existing tools into the smooth pipeline. The scalability i.e. dealing with many concurrent users may also be an issue. Then there is a whole deployment strategy to be implemented: internal servers or cloud systems, what machines, how to assure that the programmatic environment is the same as used locally. Setting access rights and logging the user’s activity, securing the data and the results must be taken care of. Not to mention the process of maintaining the code base, bug fixing, testing, releasing updates and setting the common environment for the developers. All of those aspects are crucial parts of the system and require a lot of knowledge and experience to be properly delivered.</p><p><em>JC: My take-away from the interview is that there are many exciting possibilities but one must choose the right tools for the job. Engaging an AI or DS team early in the process allows for a reality check that can save organizations time and money. Working closely with the client means devising ways to build valuable features each week during production. It’s a learning process for the subject matter experts on the client side and for the DS/AI team. They converse and educate each other to chart a course that leads to a desirable destination.</em></p><p>Thanks for reading! Follow me on Twitter <a href="https://twitter.com/_joecha_">@_joecha</a></p><p>Marcin Dubel is on Twitter as well <a href="https://twitter.com/DubelMarcin">@dubelmarcin</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23835d716f86" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/how-to-generate-ai-and-ds-use-cases-for-your-business-23835d716f86">How to Generate AI and DS Use Cases for Your Business</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Implement Intelligence Augmentation (IA) Solutions]]></title>
            <link>https://medium.datadriveninvestor.com/how-to-implement-intelligence-augmentation-ia-solutions-ded4382d8a7f?source=rss-31a888191d05------2</link>
            <guid isPermaLink="false">https://medium.com/p/ded4382d8a7f</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[intelligence-augmentation]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Joseph Cha]]></dc:creator>
            <pubDate>Wed, 30 Oct 2019 22:56:03 GMT</pubDate>
            <atom:updated>2019-10-31T16:18:40.789Z</atom:updated>
            <content:encoded><![CDATA[<p>To achieve IA implementation success <strong>plan for a partnership between human and machine, </strong>and not a replacement scenario</p><h3>TL;DR</h3><p>To achieve IA implementation success, <strong>plan for a partnership, </strong>and not a replacement, between humans and machines. Buy-in is crucial, of course. So educate your workforce about IA and plan for hero moments. Automate tasks, not jobs. Start small. And don’t over-spend on IT infrastructure. Engage the IA experts early to ensure data quality, even if you’re just starting to collect data.</p><h3>How to Implement an IA Solution: The Process</h3><figure><img alt="Depiction of the helpful AI “GERTY” from Moon (2009), directed by Duncan Jones" src="https://cdn-images-1.medium.com/max/1024/0*tbHQVo2IwgQgM_Wj.jpg" /><figcaption>Depiction of the helpful AI “GERTY” from <a href="https://en.wikipedia.org/wiki/Moon_(film)">Moon</a> (2009), directed by <a href="https://twitter.com/ManMadeMoon/">Duncan Jones</a></figcaption></figure><p>There are several keys to implementing a reliable Intelligence Augmentation (IA) solution that scales and provides maximum value. The technology for AI and IA is the same, but the difference in the end result can be traced to involvement of subject matter expert humans in the planning of the application and in the use of it (for the previous articles in the series, see the links at the bottom of this post).</p><p><a href="https://www.datadriveninvestor.com/2019/02/19/artificial-intelligence-trends-to-watch-this-year/">5 Artificial Intelligence Trends To Watch This Year | Data Driven Investor</a></p><p>Andrew Ng offered several solid tips in this area during his “<a href="https://www.deeplearning.ai/blog/andrew-ng-at-amazon-remars-2019/">Amazon re:MARS 2019 in Las Vegas, Nevada</a>” presentation. He is an <strong>AI</strong> expert, but the tips apply to <strong>IA</strong> as well. I summarize them below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/0*wyGNdX_4xEWTUjnr.png" /><figcaption>Andrew Ng presenting at <a href="https://www.deeplearning.ai/blog/andrew-ng-at-amazon-remars-2019/">Amazon RE:Mars 2019</a></figcaption></figure><ol><li><strong>Start small</strong>. Find a six-month project and do it quickly, gain momentum, and build from there.</li><li>Automate <strong>tasks</strong>, not jobs.</li><li>Combine AI and <strong>Subject Matter</strong> expertise.</li><li>Brainstorm six projects. Evaluate, then <strong>select one or two</strong> to invest in.</li></ol><p>Mr. Ng explains that there is a certain set of things that AI can do, and there is a set of activities that can help a business. Unfortunately today, only AI experts have a good sense of ‘What AI Can Do,’ and only <strong>Subject Matter Experts</strong> — people who know your business — have a good sense of what goes in the category of ‘Things Valuable to your Business.’ So Mr. Ng recommends to most companies to form <strong>cross-functional</strong> teams to where people sit down and brainstorm projects together. List out and discuss the activities or problems that fall in each category, and find the matches.</p><p>I already used Mr. Ng’s radiologist example in a <a href="https://appsilon.com/decision-making-support-systems-3-differences-between-ia-and-ai/">previous post</a>, but it bears repeating here. It’s an excellent example of how to identify use cases for AI by automating tasks, not jobs:</p><blockquote><em>I’ve found it more useful to look at </em><strong><em>jobs</em></strong><em>, which comprise many smaller </em><strong><em>tasks</em></strong><em>, and then look at the tasks that make up a job and see what’s </em><strong><em>amenable</em></strong><em> for AI automation. For example, take the healthcare industry. Radiologists do a lot of things. They read x-ray images, they also consult with patients, they do surgical planning, and mentor younger doctors… out of all of these tasks, </em><strong><em>one of them</em></strong><em> seems amenable to AI automation or acceleration — that’s why many teams including mine are building systems like these to have AI enter this task of reading x-ray images.</em></blockquote><p>I like this quote because it elucidates a process for identifying AI use cases at a company that isn’t called Google, Facebook, Baidu, etc. Any company or industry can gather the tasks that go into a given job and identify the ones that are routinizable, that are ripe for automation. I provided other examples of use cases turned into IA solutions in this <a href="https://appsilon.com/decision-making-support-systems-3-differences-between-ia-and-ai/">article</a>.</p><p>Andrew Ng <a href="https://towardsdatascience.com/ai-for-everyone-what-andrew-ng-want-to-convey-with-this-non-technical-course-in-30-points-bedaea57c81b">recommends elsewhere</a> that a company that wishes to incorporate AI into its processes do (or not do) the following. There are actually some real money-saving tips here:</p><ul><li>Don’t spend much on the IT infrastructure to collect data. Feed data as early as possible to an AI team so that they can figure out whether that collected data is useful and can change the data collection strategy.</li><li>Any problem that a human can do with 1 second of thought and for which lots of labeled data is available can be automated with supervised ML.</li><li>Automate tasks not jobs. Understand <strong>Pain Points</strong> in your business.</li><li>You can make progress even without big data.</li></ul><h3>Buy-In and Hero Moments</h3><p>In <a href="https://www.amazon.com/Augmented-Mind-Superhumans-Economic-Revolution-ebook/dp/B07DHYXMQ6"><em>Augmented Mind: AI, Humans and the Superhuman Revolution</em></a><em>, </em>Alex Bates describes a tried and true, successful dynamic between person and machine. He introduces us to two crucial milestones that ensure a successful implementation: <strong>buy-in</strong> and <strong>hero moments</strong>. He writes, “Parts of the <strong>methodology</strong> included educating end-users to gain a high-level understanding of how the AI worked and also enabling them to <strong>take ownership</strong> of AI agents that <strong>they created</strong>. This gave them a chance to have <strong>hero moments</strong>.”</p><p>How can an organization achieve AI buy-in from its workforce? It is a common challenge. A <a href="https://www.mckinsey.com/featured-insights/artificial-intelligence/ai-adoption-advances-but-foundational-barriers-remain">McKinsey study</a> found that in many companies, AI projects get stuck in “pilot purgatory” due to a failure in incorporating the tools into standard operating processes. The tools may perform beautifully in the pilot program, but human workers exclude the tools from their daily work. Why facilitate the training and installation of your replacement? So the goal becomes to foster a win-win scenario in which the human benefits from the AI making a valuable contribution. As we explored in the <a href="https://appsilon.com/decision-making-support-systems-3-differences-between-ia-and-ai/">previous article</a>, machines excel at certain tasks, while humans excel at others. If a machine alerts the human to an inventory problem, potential fraud, investment opportunity, or a security risk, and the human uses the information to the benefit of the organization, then we have a <strong>hero moment</strong>. The human should get credit for such successes of the partnership. This is Intelligence Augmentation, the “IA process.”</p><h3>Ready for Construction Phase</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/597/0*TDi0FXeYWBpMRteW.gif" /><figcaption>Construction of the USS Enterprise from <a href="https://en.wikipedia.org/wiki/Star_Trek_Beyond">Star Trek Beyond</a> (2016) directed by <a href="https://twitter.com/justinlin?lang=en">Justin Lin</a></figcaption></figure><p>Once you have decided on the use case and scope of your project, what are the best practices for implementing it? Here are some tips inspired by <a href="https://twitter.com/marekrog">Marek Rogala’s</a> <a href="https://insurancedatascience.org/downloads/Zurich2019/Lightning%20Talks%201/6_Rogala_Mark.pdf">presentation</a> at the Insurance Data Science conference in Zurich. The process applies to IA and AI projects:</p><p><strong>Data validation.</strong> The initial dataset that you use to train the IA model is important, but this isn’t a one-time event. Typically we receive new data as time goes on, and we want to update the model with the new data to teach the model-new things. <strong>Automated data verification</strong> is therefore necessary <a href="https://appsilon.com/data-quality/">so that we can immediately get alerted</a> if there are problems with the data.</p><p><strong>Interpretability</strong>. Especially important for insurance and finance sectors. Insurance professionals need to be prepared to explain their decisions to the users of their services. Recommendations from a model are only helpful if the decision-making process is <a href="https://christophm.github.io/interpretable-ml-book/">explainable</a>.</p><p><strong>Reliability and scaling.</strong> When putting a model into production we need a plan for how it will scale. We need to plan for events like spikes in usage or server outage. You can design the system to be deployed on multiple servers in the cloud so that new servers can be added at any time and there’s no single point of failure.</p><p><strong>Human augmentation and oversight</strong>. Intentionally design how humans can work together with the model and oversee its operation. The model can help with repetitive tasks, but in the end, it is the human that makes important decisions.</p><p><strong>Automated model update</strong>. For the successful implementation of an IA model, it needs to learn new things based on new data. This should be fully automated, and not a manual process. This way we avoid possible errors and ensure that updates happen frequently.</p><p><strong>User interface</strong>. It’s important to match your state-of-the-art IA model with a user interface that the end-users can employ in their tasks without friction. Depending on the use case, this can be an API called by other systems or a <a href="https://appsilon.com/how-we-built-a-shiny-app-for-700-users/">human-in-the-loop <strong>Shiny</strong> <strong>dashboard</strong></a>.</p><h3>Final Thoughts</h3><p>It’s doubtful that IA will out-hype AI in the near future, but maybe it should. An IA approach can help non-software professionals — which represents most of the humans in the world — to visualize opportunities for automation and decision-support. While the emergence of a General Artificial Intelligence is still a ways off, the opportunities for competitive advantage and performance enhancement through human intelligence augmentation systems are very real in the present.</p><p>This concludes the Intelligence Augmentation series. Here are the previous posts in the series:</p><p><a href="https://appsilon.com/decision-making-support-systems-1/">What Is Artificial Intelligence?</a></p><p><a href="https://appsilon.com/decision-making-support-systems-2/">What Is Intelligence Augmentation?</a></p><p><a href="https://appsilon.com/decision-making-support-systems-3-differences-between-ia-and-ai/">Differences between IA and AI</a></p><p>Thanks for reading! Follow me on Twitter <a href="https://twitter.com/_joecha_">@_joecha_</a></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ded4382d8a7f" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/how-to-implement-intelligence-augmentation-ia-solutions-ded4382d8a7f">How to Implement Intelligence Augmentation (IA) Solutions</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Differences between Intelligence Augmentation and Artificial Intelligence]]></title>
            <link>https://medium.com/@joseph_cha/the-differences-between-intelligence-augmentation-and-artificial-intelligence-23c055a2c82f?source=rss-31a888191d05------2</link>
            <guid isPermaLink="false">https://medium.com/p/23c055a2c82f</guid>
            <category><![CDATA[intelligence-augmentation]]></category>
            <category><![CDATA[decision-support]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Joseph Cha]]></dc:creator>
            <pubDate>Wed, 23 Oct 2019 14:46:19 GMT</pubDate>
            <atom:updated>2019-10-31T12:48:10.136Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>Moving away from “<em>Terminator”</em> scenarios and towards “<em>Interstellar”</em></strong></h4><p>In previous posts, we looked at the <a href="https://appsilon.com/decision-making-support-systems-1/">definition of Artificial Intelligence</a> (AI) and the <a href="https://appsilon.com/decision-making-support-systems-2/">definition of Intelligence Augmentation</a> (IA). So, what are the differences between the two? Intelligence Augmentation has always been concerned with aiding human decision making and keeping humans-in-the-loop, whereas the AI endeavor seeks to “<a href="https://www.stateof.ai/">build machines that emulate and then exceed the full range of human cognition</a>, with an eye on replacing them.</p><p>Rob Bromage puts it well in “<a href="https://www.hcamag.com/au/news/general/artificial-intelligence-vs-intelligence-augmentation/152265">Artificial intelligence vs intelligence augmentation</a>”: “AI is an autonomous system that can be taught to imitate and replace human cognitive functions. To put it simply, the machine completely replaces human intervention and interaction. <strong>IA</strong>, on the other hand, plays more of an assistive role by leveraging AI to <strong>enhance</strong> human intelligence, rather than replace it.”</p><p>In “<a href="https://medium.com/datadriveninvestor/augmented-intelligence-not-artificial-intelligence-is-the-future-f07ada7d4815">Augmented Intelligence, not Artificial Intelligence, is the Future</a>” Aaron Masih writes that “While the underlying technologies powering AI and IA <strong>are the same</strong>, the goals and applications are fundamentally different: AI aims to create systems that run without humans, whereas IA aims to create systems that make <strong>humans</strong> better.”</p><figure><img alt="Astronaut and robot cooperate in the film “Interstellar.”" src="https://cdn-images-1.medium.com/max/820/0*aODgUbffkO86DwBw.png" /><figcaption>A beneficial partnership between human and machine is depicted in <a href="https://slate.com/technology/2014/11/tars-the-interstellar-robot-should-be-the-future-of-artificial-intelligence.html">Interstellar</a>.</figcaption></figure><p>IA systems are able to exceed <strong>system boundaries </strong>(the parts of an environment not covered by an AI model’s data inputs) because of the “unsung heroes of the AI revolution” — humans! AI systems can only work with the datasets that they’re trained on and that feed them. <strong>Subject matter expert humans</strong> excel in applying context and intuition to problems. Humans can grasp and explain <a href="http://bayes.cs.ucla.edu/WHY/"><strong>causality</strong></a>. <a href="https://www.amazon.com/Augmented-Mind-Superhumans-Economic-Revolution-ebook/dp/B07DHYXMQ6">Alex Bates</a> recalls from his experience running a an AI/prescriptive maintenance firm that “…what was remarkable about the human process engineers and maintenance engineers at these plants …were all the clues they incorporated into their assessments of equipment failure, allowing them to identify what was failing and what to repair. Where they struggled, and where the AI helped, was in making sense of all the massive amount of sensor data coming off the equipment.”</p><p>Another major difference? To state the obvious, one receives much more attention than the other. Even the casual industry observer of the industry that <strong>augmenting human intelligence</strong> represents a tiny fraction of the total research and development devoted to Artificial Intelligence. The lack of investment in the area seems like a missed opportunity.</p><p>So the AI approach seeks to minimize or replace the role of humans, while an IA solution seeks to amplify the abilities and performance of the humans that participate in a given activity. The IA approach benefits from the human ability to think outside system boundaries. But despite the lack of investment and press attention on Intelligence Augmentation, businesses and other organizations continue to benefit from IA applications…</p><h3>Examples of IA at work today</h3><figure><img alt="Filip Stachura presents “Why Shiny” at useR Toulouse 2019" src="https://cdn-images-1.medium.com/max/480/0*hhZ0HOLGccPqWriH.jpg" /><figcaption>CEO of Appsilon Filip Stachura at <a href="http://ceo%20of%20appsilon%20filip%20stachura%20at%20user%21%20toulouse%202019/">useR! Toulouse 2019</a></figcaption></figure><p>I asked <a href="https://twitter.com/filipstachura">Filip Stachura</a>, CEO of Appsilon Data Science, How much of what you do as a company qualifies as Intelligence Augmentation?</p><blockquote>Most of what we do qualifies as IA. Decision support systems can be more realistic in many business cases. Even a recommendation engine application is a sales-support decision system, not the AI ​​that sells itself.</blockquote><p>“For one firm we did the following: Prices for products that our client sells change frequently and depend heavily on negotiations. When a client requests a product, the salesperson opens the application that we built and fills in the client and product name. They then see the history of deals made with the client and <strong>suggested</strong> prices/discounts for the product. The prices vary depending on the segment of the business, category/size of the client and the size of the order. Ultimately, the human salespeople make the pricing decision, but the application provides the most up to date information to optimize decision-making. It’s sales-support, not salesperson replacement.”</p><p>“Here is another example. What if a manager has to optimize the usage of hundreds of varying models of cleaning machines that are based in different locations in a region? Each location has its own logistical needs and have staff of varying levels of expertise in operating the machines. How do you optimize performance and cost with such diverse conditions? Is the answer to eliminate the manager in charge, or is it to augment the manager’s capabilities with a decision-support system? Obviously it’s the latter, since a fully automated management system would have great difficulty in assessing non-standard situations and communicating with the various teams at the various locations.”</p><p>Appsilon Data Science co-founder <a href="https://twitter.com/D_Rodziewicz">Damian Rodziewicz</a> added that “We also worked with <a href="https://twitter.com/kenbenoit">Dr. Ken Benoit</a> and his team from the London School of Economics to make their <a href="https://quanteda.io/"><strong>Quanteda</strong> text analysis tool</a> available to a much greater number of social scientists and practitioners from other fields, including medicine and law. Now the user doesn’t need to know the R programming language in order to make use of the powerful Quanteda R package. A human researcher can use the Quanteda package to quickly evaluate language and social trends from millions of documents, truly extending the human’s research domain by many orders of magnitude.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hotrZ500krhFV6xB.jpeg" /><figcaption>The man and the machine: Dr. Ken Benoit and the <a href="https://quanteda.io/index.html">Quanteda</a> text analysis tool</figcaption></figure><p>Here is another example, and I have transcribed and hijacked it from the Andrew Ng “<a href="https://www.deeplearning.ai/blog/andrew-ng-at-amazon-remars-2019/">Amazon re:MARS 2019 in Las Vegas, Nevada</a>” presentation. He gave it as an example of AI, but I think it’s really an example of IA.</p><figure><img alt="Andrew Ng presents at re:MARS 2019. His slide says “Automate Tasks, Not Jobs.”" src="https://cdn-images-1.medium.com/max/789/1*KFRkvVGQ6em8PXCj33Sjow.png" /><figcaption>Andrew Ng at re:MARS 2019</figcaption></figure><blockquote><em>Radiologists do a lot of things. They read x-ray images, they also consult with patients, they do surgical planning, and mentor younger doctors… out of all of these tasks, one of them seems amenable to AI automation or acceleration — that’s why many teams including mine are building systems like these to have AI enter this task of reading x-ray images</em></blockquote><p>Does anyone really want to <strong>replace</strong> radiologists with machines at this time? With human lives at stake? Probably not. But a machine-assisted radiologist? That is interesting. After the first versions are released, radiologists can work with engineers to teach the machines to do a better job in finding problems. And eventually there can be a feature set that allows the radiologists to teach the machines directly, without the constant participation of the engineers. And training data from all over the world can be shared to further optimize results.</p><h3>When is IA superior?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*DviP4mOHsqglIB7B.jpg" /><figcaption>Another beneficial human and machine partnership depicted in “<a href="https://www.imdb.com/title/tt9788508/">Lucky 13</a>.”</figcaption></figure><p>In short, IA is superior when…</p><p>…only a <strong>limited</strong> amount of labeled data, or only unlabeled data exists for a given task</p><p>…a task requires <strong>empathy</strong> and//or negotiation between humans</p><p>…a task requires a notion of <strong>causality</strong>, not just correlation</p><p>…<strong>system boundaries</strong> exist. Crucial data cannot be captured by sensors. Data input doesn’t cover the entirety of the problem.</p><p>….a problem exists within a <strong>regulated environment</strong> in which machines still struggle with decisions, such as in medicine and surgery</p><p>…a task is so critical that only a human can make the final decision</p><p>…a task requires more than <a href="https://towardsdatascience.com/ai-for-everyone-what-andrew-ng-want-to-convey-with-this-non-technical-course-in-30-points-bedaea57c81b">2–3 seconds</a> of human thought</p><p>The above criteria probably describe most business and research problem scenarios. Consider moving away from the approach of “how do we replace our staff with Artificial Intelligence agents,” and instead move towards “what repeatable, routinizable task can we automate in order to free up time for our human teammates? How do we unleash more intuitive and brilliant ideas by increasing their bandwidth? How do we prevent problems in our facilities by partnering machines and humans?”</p><p>Thanks for reading! Follow me on Twitter <a href="https://twitter.com/_joecha_">@_joecha_</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23c055a2c82f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>