<?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 Minh Dao on Medium]]></title>
        <description><![CDATA[Stories by Minh Dao on Medium]]></description>
        <link>https://medium.com/@minh.dao_80561?source=rss-4774d7f48bb4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*bJm0pKxw18LOSAoInSkhnQ.jpeg</url>
            <title>Stories by Minh Dao on Medium</title>
            <link>https://medium.com/@minh.dao_80561?source=rss-4774d7f48bb4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 16:06:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@minh.dao_80561/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[What the Future of AI-enabled Products Will Need]]></title>
            <link>https://medium.com/@minh.dao_80561/what-the-future-of-ai-enabled-products-will-need-50080e874865?source=rss-4774d7f48bb4------2</link>
            <guid isPermaLink="false">https://medium.com/p/50080e874865</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Minh Dao]]></dc:creator>
            <pubDate>Tue, 30 Jul 2019 13:01:01 GMT</pubDate>
            <atom:updated>2019-07-30T13:47:31.605Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Every day we hear how artificial intelligence has accomplished bigger and bolder feats: </em><a href="https://www.youtube.com/watch?v=D5VN56jQMWM&amp;source=post_page---------------------------"><em>Google Duplex</em></a><em> can call a restaurant and book you a reservation. DeepMind’s </em><a href="https://www.youtube.com/watch?v=UuhECwm31dM&amp;source=post_page---------------------------"><em>Alpha Star</em></a><em> has defeated the world’s top Starcraft video game players. Yet AI still hasn’t fully infiltrated every aspect of our lives as some of the hype has led us to believe — where are our self- driving taxis?</em></p><p><em>We still have major strides to make in the realm of AI technology before we can expect to see it everywhere. More specifically, before AI can become ubiquitous, we need to better trust it.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XGdvR91la4akBt6V57lvwA.jpeg" /></figure><p>Can you imagine a future where you’re travelling in a completely autonomous airplane with no pilot? How would you feel? Would you trust an airline that claimed it could get you from point A to point B without a human in the cockpit?</p><p>Trusting autonomous vehicles and other types of AI in high-risk situations seems like a fantasy reserved for sci-fi set in the future — yet planes have been running on autopilot for decades and can now even <a href="https://techcrunch.com/2019/07/05/watch-a-plane-land-itself-truly-autonomously-for-the-first-time/?source=post_page---------------------------">land completely autonomously:</a></p><p><strong>Small plane automatically lands itself</strong></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FtPFC7Ki2JT0%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DtPFC7Ki2JT0&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FtPFC7Ki2JT0%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/fb0c3620cfaff6095cba687764c25361/href">https://medium.com/media/fb0c3620cfaff6095cba687764c25361/href</a></iframe><p>Whether it’s with good judgement or not, people are proving that they are more and more willing to (in some cases) trust AI — <strong>as long as they receive adequate value from a service or product in return</strong>. Case in point: the number of reports we see of people driving their Teslas on auto-pilot while on their phone or even <a href="https://jalopnik.com/heres-another-dummy-apparently-sleeping-at-the-wheel-of-1833123703">completely asleep at the wheel</a>.</p><h3>SETH BLAKE on Twitter</h3><p>Dude is straight snoozing going 75mph on the interstate, letting his @Tesla do the work. 😳😴 https://t.co/RQD2LBSnGh</p><p><strong>Chances are you’re already implicitly trusting AI</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/820/1*cikHE7DcmaJ3b3vqvJdK4Q.png" /></figure><p>When you’re traveling through a busy city and you pull out your AI-enabled GPS navigation app, you’re trusting it will tell you <em>exactly</em> where you need to go, quickly and safely. Fortunately, the chances of the navigation being wrong are low — at worst, you may run a little late or accidentally arrive at the wrong address. These scenarios are of relatively low consequence, and are well worth the value exchange of removing considerable cognitive load (no more snap decisions while navigating in a hurry) — all for free.</p><p>But what about in higher-risk scenarios? Would you blindly trust AI to make decisions on your behalf about your health? If we are to fully embrace and trust AI products as applied to more consequential parts of our lives, it’s not likely we will do so blindly.</p><p>We will need to <strong>develop new ways to interact with these AI systems and ultimately build trust.</strong></p><h3><strong>Trust and social acceptance — the barrier to widespread adoption</strong></h3><p>In his book The<a href="https://www.amazon.ca/Inevitable-Understanding-Technological-Forces-Future/dp/0525428089?source=post_page---------------------------"><em> Inevitable</em></a><em>,</em> Kevin Kelly shares three main conditions that must be in place in order for a new technology to become successful and widespread:</p><p>1. Consumable technology: It’s technically feasible, available to the masses and achieved some level of product-market fit.</p><p>2 . Ubiquitous and affordable technology: The tools to create software and hardware are easy to obtain and affordable.</p><p>3. Social and cultural acceptance: This happens when a technology is widely used and accepted as the new status quo.</p><p>Let’s look at the internet as an example. Once an invention of the military, the internet later found its place as a communication tool. More affordable internet providers, home computers and better coding languages made the internet more accessible to the masses. Today, you’re on the internet on your phone in public anytime (and all the time). The UN considers internet access a <a href="https://www.pastemagazine.com/articles/2016/07/the-un-declares-internet-access-a-basic-human-righ.html">basic human right</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/699/1*rJf1fCKSWSkUVIrfLAvV8w.jpeg" /><figcaption>GPU computing performance expected to be 1000x by 2025</figcaption></figure><p>AI technology is already here and starting to make its way into the hands of consumers — the acceleration of affordable hardware such as GPUs and tools to create AI is moving at an unprecedented rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/1*cvcXn5K6Xt_vjV9uL7kJ8w.jpeg" /><figcaption>NVIDIA Jetson Nano AI computer available for purchase for $99 in March 2019</figcaption></figure><p>But that third factor that Kelly cites — social and cultural acceptance — has yet to become a reality. And it’s not likely to happen soon if we do not <strong>consciously develop methods for building trust.</strong></p><p>In addition to building trust with users and easing adoption, these methods could also keep creators honest in the creation of responsible AI that does not exploit its users.</p><h3><strong>Key factors for AI-enabled product adoption</strong></h3><p><strong>1 — Explainability</strong></p><p>Explainable AI refers to techniques in artificial intelligence which clearly demonstrate to humans how decisions have been made by AI. It contrasts with the concept of the “<a href="https://en.wikipedia.org/wiki/Black_box?source=post_page---------------------------">black box</a>” in machine learning where even their designers cannot explain why the AI arrived at a specific decision.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FgB_-LabED68%3Fstart%3D88%26feature%3Doembed%26start%3D88&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DgB_-LabED68&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FgB_-LabED68%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/2a1270f4f7fa51baad3c828304a0bbe4/href">https://medium.com/media/2a1270f4f7fa51baad3c828304a0bbe4/href</a></iframe><p>As our products begin to make higher-risk decisions on our behalf, consumers are likely to demand a peek into the black box.</p><p>As digital products become increasingly complex, our methods of showing what’s going on behind the scenes will need to remain transparent. These complex models and datasets can be made up of hundreds of millions of data points (for example: number of users, what each user has viewed, which movie actors/actresses they prefer, which genre of movies they prefer). All of these disparate signals are then analyzed to offer a simple recommendation or match score. Even researchers don’t always understand exactly what’s in all parts of a neural network. This is like cutting open your brain in order to try to understand why you chose to eat tacos for lunch — the decision can not be easily traced and understood. If we are to have autonomous products making important decisions on our behalf both the creators and end users will need to understand the decisions it’s making and have insights into its inner workings.</p><p><strong>2 — Accountability &amp; Traceability</strong></p><p>When you visit a doctor at the hospital or go in for surgery, your health professional is accountable. All practicing doctors take the Hippocratic oath to do no harm and there are consequences when they make mistakes. So what happens when one day when you get treated by an AI-enabled machine? Is it accountable the same way a surgeon is? Or a civil engineer?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/660/1*mejjs8q2xKxhVnP8S0ohkw.jpeg" /></figure><p>AI is already outperforming doctors at diagnosing brain tumour and predicting hematoma expansion. In 2018 AI systems from <a href="https://www.latimes.com/world/la-fg-china-ai-20180706-story.html?source=post_page---------------------------">BioMind</a> outperformed human physicians handily on both types of neuroimaging recognition. More recently, in 2019, a team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Massachusetts General Hospital (MGH) were able to <a href="http://news.mit.edu/2019/using-ai-predict-breast-cancer-and-personalize-care-0507">predict breast cancer up to five years in advance</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6P8kJmJA4OoNVEhcIwBHuw.png" /></figure><p>Explainability could be one way of creating accountable AI products. The application could cross-check potential biases, such as modifying the dataset of the model and comparing those results. Humans could query the AI with natural language questions: “How many times have you been wrong in the past year?” “What information/data are you using in order to make that match?”</p><p>The algorithm or model needs to be wrapped with the ability to explain itself and defend its answers (and how it came to conclusions). This might be something which is submitted with the result in order to prove its credibility, viability and accuracy. An example of this is <a href="https://www.theverge.com/2018/2/27/17054740/palantir-predictive-policing-tool-new-orleans-nopd?source=post_page---------------------------">predictive policing technology</a> field where the algorithm does not necessarily need to defend its prediction.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F7lpCWxlRFAw%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D7lpCWxlRFAw&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F7lpCWxlRFAw%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6b479337a559990ae61e193537f441a8/href">https://medium.com/media/6b479337a559990ae61e193537f441a8/href</a></iframe><p>We can’t reprimand a machine like a human (perhaps a demotion to an overqualified soda machine? ;), so AI practitioners will need to be creative about baking accountability into their AI products. The researchers, data scientists and engineers who are creating the technologies and AI products must understand and empathize with the people who will be using them.</p><p><strong>3 — Trustability</strong></p><p>All of these paths lead to building trust: developing methods for collaborating and interacting with AI with ease and confidence.</p><p>How do you determine whether you can trust a person you just met? You might ask questions, see how the other person responds, add context to their motivations and gauge whether they are acting with full transparency.</p><p>We can build trust in AI in much the same way we build trust amongst ourselves: Building trust can be done similarly to the way humans build trust — by building relationships.</p><p>An example we can use is the Boeing 737 Max 8 airplane.</p><p>While its not certain what caused the malfunction in two airplanes, reports suggest a <a href="https://boeing.mediaroom.com/news-releases-statements?item=130402&amp;source=post_page---------------------------">software problem</a> was the cause; the automated stall-prevention system was malfunctioning. In both cases, the airplanes were getting incorrect sensor data and fighting against the pilot trying to pull up the plane. It’s also been stated that the pilots were not adequately aware of the new stall-prevention system in a plane which looked exactly like the airplanes predecessor that did not have the automated system.</p><p><strong>In order for AI machines to build trust from its users, there needs to be strong two-way communication. </strong>I’m not talking about the kind of interaction you have with Siri — I’m talking about the interaction you’d have with a trusted friend or colleague, a two-way dialogue and exchange of important information. AI needs mechanisms for empathizing with humans, and being flexible to our needs and wants. It should be able to read the cues we put out into the world (it’s not only what you say but also <em>how</em> you say it).</p><p>For example, if we are to have augmented vehicles in the future, they could adapt to different scenarios:</p><ul><li>Let’s say you’re driving through thick rush hour traffic. Your car speaks to you in a calmer tone, only pushing important notifications in order not to divide our attention and highly-sensitive to any potential dangers.</li><li>Now let’s say you’re driving on a long, winding road. The vehicle allows for more aggressive driving and for you to push its performance. This could be done by actively communicating the driver, watching his/her physical response/abilities on the road.</li></ul><p><em>“Frank I’m noticing increase in speed and acceleration. To get better throttle response I’ve gone ahead and activated sports mode. The sensors tell me the road is quite wet and an increase in oversteer. Did you want me to turn up the traction control by 20%?”</em></p><p>These are hypothetical examples. The truth is there is no silver bullet when it comes to how we will be able to trust an AI. It will depend on context, the importance of that task and the person. For this reason the AI needs to be flexible and able to interpret the signals and feedback in the environment.</p><h3>An increase in machine automation and complexity results in the need for a high level of explicit trust</h3><p>Like human relationships, we have a trust battery with machines. Only as more trust is established can we then allow ourselves to rely on machines to make more important and risky decisions without the need for a human in the loop.</p><h4>Required trust in AI is directly proportional to the automation it provides</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JMdOsh1Fb53lRlr_0ZulLg.jpeg" /></figure><p>The diagram above illustrates how the need for trust in an AI system, service or product is correlated to the level of automation. When you are able to trust the output, you are able to achieve riskier and more complex decision-making or actions (listed along the y-axis). Product recommendations on Amazon are one example of a low-risk automated task, where you may only need a basic explanation. Making a high-risk decision (moving up the y-axis) such as recommending knee surgery based on medical imaging requires a higher level of trust and more robust reasoning and feedback for the patient.</p><h3>Trust is created by us for us</h3><p>In my <a href="https://medium.com/@minh.dao_80561/the-future-of-ai-products-complex-for-creators-but-simple-for-users-9492d5909459">previous article</a> I talked about the shift in complexity and effort from humans onto the product and machine. If we are to be successful in creating strong and effective AI-enabled products we will need to demand to see transparent decision making and moreover trustability of AI in our products.</p><p>If we are truly going to live in a world where AI products are going to enrich our lives and allow us to focus on more important activities, we need to trust them explicitly and not blindly. Governments, companies and the product creators ultimately have the onus of giving us great technologies and products but also to steer us into an ethically strong future.</p><p>Thanks for the read.</p><p>This is only my second blog article in 12 months. If you like this one and would like to see more, please leave a clap or comment. It will motivate me to get my ass in gear to write more. The plan is to publish my next article, Human and AI Collaboration, before the end of the year.</p><p>If you’d like to go more in-depth on trustability in AI, there’s a great article by Bahador Khaleghi called —<a href="https://www.elementai.com/news/2018/breaking-down-ais-trustability-challenges"> Breaking Down AI’s Trustability Challenges</a></p><p>Author: Minh Dao<br>Editor: Amanda Durepos<br>Contributors: Santiago Salcido, Chrystel Black<br>Thanks also to: Nathan Schucher, Thomas Boquet, Patrick Steeves</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=50080e874865" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Future of AI Products: Complex for Creators but Simple for Users]]></title>
            <link>https://medium.com/@minh.dao_80561/the-future-of-ai-products-complex-for-creators-but-simple-for-users-9492d5909459?source=rss-4774d7f48bb4------2</link>
            <guid isPermaLink="false">https://medium.com/p/9492d5909459</guid>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Minh Dao]]></dc:creator>
            <pubDate>Thu, 11 Oct 2018 11:01:02 GMT</pubDate>
            <atom:updated>2018-10-12T00:46:51.008Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l-VS_3FkiqlCiexBWVf-Ww.jpeg" /></figure><p>Think back to life a decade ago: before you could query Siri, stream a movie on Netflix, or order a car with Uber. We take these conveniences for granted today; it’s hard to imagine life without them.</p><p>But try and imagine what the world could look like a decade from now and you’ll be even more hard pressed. We’ve already grown accustomed to having artificial intelligence in our pockets and in our homes that have deeply augmented our connection and interaction with the world, yet it’s clear that we’re only in the beginning stages of incorporating AI into everyday products.</p><p>Increasingly, digital products are designed to require less human inputs and make decisions on our behalf. <strong>The effort is shifting from the user to the product </strong>— so what does this mean for future AI-enabled products and the way they are designed?</p><h3>AI product effort triad</h3><p>Any product requires a certain amount of power, complexity and processing to produce and use. For every digital product, this effort can be broken into three components:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WRFL4-rakcWdRVW6gaXIhQ.png" /></figure><ol><li>The effort exerted by the <strong>product</strong> itself (AI-enabled features that reduce effort and increase value for the user)</li><li>The effort exerted by product<strong> creators</strong> (the engineers, designers and scientists who build these features)</li><li>The effort required by end <strong>users</strong> to actually use the product (a company employee or consumer like your grandma, for example)</li></ol><p>Generally, a product creator aims to reduce the amount of effort and increase value for their end user while ensuring that the effort required to build these products is efficient and streamlined. In other words, we strive to create products that <strong>reduce effort for humans and shift this responsibility to AI-enabled machines.</strong></p><p>To understand where we’re going, let’s first look at how we got here.</p><h3>More complex designs for a better user experience</h3><p>For many years, products largely had simple design and basic functions. Take a hammer for example: blunt, easy to use, without much of a learning curve. A hammer has straightforward affordances that teach us how to use it: hold the hammer on one end and pull nails with the other.</p><p>Fast forward to the 1980s, and creators are introducing more complexity into their designs. A <strong>VCR</strong> for example has many components (the machine itself, cables connecting it to our television, tapes to put in it, the remote control) and requires a high learning curve in order to operate. I dare you to explain to someone how to schedule their VCR to record a TV program — not easy.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FrDdvUxyuT2c%3Fstart%3D45%26feature%3Doembed%26start%3D45&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DrDdvUxyuT2c&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FrDdvUxyuT2c%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/1d27e3a33abd56660e1a2ff5abe06dbd/href">https://medium.com/media/1d27e3a33abd56660e1a2ff5abe06dbd/href</a></iframe><p>But as we gain access to more processing power and more complex technology, products are offering more and more value to consumers.</p><p>Creating a product like <strong>Netflix</strong> requires a lot of top-tier engineering, developers and data scientists, but the end result is massive value for millions of users worldwide. The product takes on the bulk of the effort by using machine learning to determine your taste and cater recommendations to you. This is no doubt valuable to users; Netflix reports that over <a href="http://www.businessinsider.com/netflixs-recommendation-engine-drives-75-of-viewership-2012-4">75% of what their users watch comes</a> from a recommendation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kwYGPRur4rbu6qE5zVUNSA@2x.png" /></figure><p>No more searching through your VHS collection, rewinding a tape, traveling to Blockbuster only to circle inside for 30 minutes with your friend, struggling to settle on a movie you both want to watch. Instead, you can pick from thousands of (curated) options and play them on demand — all within a few seconds. <strong>This delivers remarkable value to the end user for such little cost, time and effort.</strong></p><p>AI-powered products can offer the best of both worlds: products that are as simple as a hammer but as feature-rich as Netflix.</p><h3>Future product trends</h3><p>Product creators are already using AI to reduce the workload of users, but what might the full extension of this look like?</p><p>Thanks to your new smart home thermostat and lighting system, the A/C turns on as you’re on your way home from work, and entry lighting flips on as you open the door. When you leave for vacation, integration with your GPS signals that your smart home should go into “vacation mode,” where it will reduce power (but still flip on lights occasionally according to your regular patterns to deter burglars).</p><p>After your vacation, you’re back at work and trying to get to inbox zero. Imagine an email client that can sort, filter, and prioritize your messages… and even answer some of them!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gQgciY2E6OZT-eLsh1J_4A.png" /></figure><p>Gmail is already using ML to make answering emails easy, just not quite to the extent detailed above.</p><p>While it’s unclear what technologies will actually exist in the future, certain trends feel like safe bets.</p><h3>1. Fewer interfaces</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2G3lH9ztmF-FFX9Tm9svFw.jpeg" /></figure><p>If we strip down the necessary inputs and outputs required by a rider in a driverless car, the only input the vehicle would really need is to know where would you like to go. The only output needed? ETA to the destination. All of this in theory could take place via spoken language:</p><p>“Where do you need to go?”</p><p>“The dentist.”</p><p>“There’s a bit of traffic on Mont Royal Avenue. ETA: 14 minutes.”</p><p>There should be no need for any other interaction from the user. If absolutely necessary, additional interaction could also take place via simple conversation Natural Language Processing (NLP).</p><h3>2. Flattened learning curve</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BR72CbEDLyihwU5er6QM2w.jpeg" /></figure><p>Imagine AR (augmented reality) safety glasses that a worker puts on while on a factory floor.</p><p>The AR could on-board and teach the user what the tasks are step-by-step and offer help when asked (“Where does this screw go?” “The screwdriver torque setting should be ‘5’”). In addition to this, the glasses could even predict error or potential accidents.</p><p>Other future applications could be: learning to play a sport, visiting a new city, or instructions on using a product.</p><h3>3. Highly-customized product experiences</h3><p>Often, products are built to deliver the most value possible to the majority — mass producible and scaled for millions of users — but doing this limits the ability to offer unique product experiences to everyone.</p><p>AI-enabled apps and products will no longer require us to make hard trade-offs on which users and use cases to build for. Machine learning is enabling the creation of products that adapt to every individual’s unique needs and preferences; think of our Netflix example from earlier. This kind of hyper-personalization and customization has already started to shift customer expectations about the products they use.</p><p>The full extension of this is exciting to think about; envision a future wherein app development is not limited to those who learn specific languages, but could be done using voice commands from people with varying backgrounds and skillsets.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F_oet4GOzcRQ%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D_oet4GOzcRQ&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F_oet4GOzcRQ%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/ab5bcff784850051e1df70348132ddf7/href">https://medium.com/media/ab5bcff784850051e1df70348132ddf7/href</a></iframe><h3>4. Increasing the human output bandwidth</h3><p>Elon Musk often discusses human input and output constraints. Specifically while our eyes can pick up tons of information in our surroundings or from an image, our output capabilities are quite limited — usually to your mobile phone and your two thumbs.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FZrGPuUQsDjo%3Fstart%3D97%26feature%3Doembed%26start%3D97&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DZrGPuUQsDjo&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FZrGPuUQsDjo%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/3ea01e5cddaf469c8e678f5a86cc2b02/href">https://medium.com/media/3ea01e5cddaf469c8e678f5a86cc2b02/href</a></iframe><p>As natural language understanding (NLU) and vision models continue to improve, we should expect the objects and products around us to be able gather more information and instruction other than just from our two thumbs and camera on a smartphone. Voice apps such as Google Assistant have allowed us to more efficiently search for information and give commands without even picking up your phone. This will continue to increase in the future as we seek more convenience, continue the trend of on-demand culture and our surroundings begin to be more AI-enabled.</p><p>Checkout-less grocery stores like Amazon Go are already a reality — it’s not too farfetched to imagine going to your local grocery store and asking an AI-assistant “Where do I find Avocado’s?” “Is this Avocado ripe?”</p><h3>Summary</h3><p>While these trends are mere bets, one thing is certain about the future of digital products:</p><p>The product should be able to answer why it chose one action over another. This can help create trust through transparency. We should be able to ask any AI what data and information about us is being used to make decisions and recommendation on our behalf.</p><p><strong>The less users are consulted on actions or tasks completed on our behalf, the more we need to build trust with our AI products.</strong></p><p>When we think of digital products of the future they will be easy to use yet full of complexity, features and value, with the effort required by the user passed off into the hands of the product creators. We must be prepared to offer the <strong>education </strong>and <strong>transparency</strong> needed for this kind of paradigm shift.</p><p>Follow me to be notified when I publish my next article: <strong>AI products— transparent decision making</strong></p><p>Author: Minh Dao<br>Editor: Amanda Durepos<br>Contributors: Chrystel Black, Erika Khanna, Patrick Lazarus, Issam Laradji, Simon Hudson, Nathan Schucher</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9492d5909459" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>