<?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 Guy TONYE on Medium]]></title>
        <description><![CDATA[Stories by Guy TONYE on Medium]]></description>
        <link>https://medium.com/@gtonye?source=rss-11382a5b2c64------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*HDfbvoQNnRFFuvNpQb1OgQ.jpeg</url>
            <title>Stories by Guy TONYE on Medium</title>
            <link>https://medium.com/@gtonye?source=rss-11382a5b2c64------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 05:13:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@gtonye/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[APIs — All You Need To Know As A Conversation Designer]]></title>
            <link>https://medium.com/voice-tech-global/apis-all-you-need-to-know-as-a-conversation-designer-2980a6dcccd8?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/2980a6dcccd8</guid>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[conversation-design]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Wed, 22 Sep 2021 23:33:02 GMT</pubDate>
            <atom:updated>2021-09-23T03:52:12.447Z</atom:updated>
            <content:encoded><![CDATA[<h3>APIs — All You Need To Know As A Conversation Designer</h3><p>It’s lunch, and you’re hungry. Now.</p><p>You ask your favourite conversational app: “Give me a list of the restaurants close to me.”</p><p>Seconds later, you hear:</p><blockquote>“I found a few restaurants nearby: Maya Bay Thai snack bar 0.4 miles away on Carlton St, Gushi Japanese Street food 0.3 miles away on Gerrard St, and Ackee Tree Dynamite Jamaican 0.2 miles away on Queen St. Would you like to know more about one of them?”</blockquote><p>This <em>bot</em> response illustrates several challenges for Conversation Design when working with <strong>dynamic data</strong>, or information that changes based on circumstances (such as your current location):</p><ul><li>Imagine that you are in a neighbourhood with many (10+) options. How should the <em>bot</em> surface so many results to the user?</li><li>What were the criteria used to choose the three restaurants in the response? And how does the <em>user</em> access the rest of the options?</li><li>How should the information be presented to the <em>user</em> to help them make a choice?</li></ul><p>In this article, we will review where the data that applications use comes from, and how Conversation Design frameworks can be used to craft a good <em>user</em> experience with dynamic data.</p><figure><img alt="Illustration for the post. The image show a hand plugging a cable near the title of the article." src="https://cdn-images-1.medium.com/max/1024/0*O_x4PAHDeP-P0YtU" /></figure><h3>What are APIs?</h3><p>A swimming pool reservation app should be able to handle simple requests, such as “I want to book an appointment this week”. But to do this, the application needs to know a lot of information, such as what reservations are available and when.</p><p>Without a bot, you might:</p><ul><li>Call the swimming pool management every time you need to make a reservation</li><li>Book a recurring appointment daily or weekly</li></ul><p>But this means the office managing the swimming pool must have a person available to answer your calls, which stops them from offering 24 hours / 7 days a week service.</p><p>Or life happens and that ongoing booking you made needs to be canceled one or two days, or changed manually.</p><p>This example is a simple demonstration of why APIs exist.</p><p><strong>API stands for Application Programming Interface. It’s a combination of protocol (similar to instructions) and software that allow computer systems to interact with each other.</strong></p><p>The main use case for an API is to allow one central information technology (IT) system, to expose their data in a secure way for other IT systems to manipulate in real-time.</p><p>In our swimming pool example, the data is the list of available dates and times, or a list of open <em>slots</em>. This data is owned by the swimming pool’s management information system. Any authorized applications that want to book or alter appointment <em>slots</em> can do so by accessing the data with an API.</p><p>Although they can have many shapes and forms, APIs are usually formatted to look like URLs. For example, the swimming pool API URL might look like: “https://api.swimmingpoolmanagement.io”.</p><p>For some APIs, it’s possible to go to the URL, like you would do with any link. You would see the data in your browser, which might look like <a href="https://binaryjazz.us/wp-json/genrenator/v1/story/25/">this music story generator</a>. For other APIs, you need to use a process to access the data. This often requires some programming or the use of no-code tools, such as <a href="https://zapier.com/">Zapier</a>.</p><p>This is where the IT systems come in. They need to have certain credentials that can securely access the information in the API. Once accessed, the information shared by the API can then be used by an app to accomplish tasks, such as booking a reservation at your local swimming pool.</p><h3>How are APIs used in Conversational AI?</h3><p>Any application that uses an API can perform four main tasks with the data available:</p><ul><li><em>Retrieve</em> data</li><li><em>Create</em> new entries</li><li><em>Update</em> existing data</li><li><em>Delete</em> entries</li></ul><p>If we look at our swimming pool reservation app example, the API can be used to access data, and then allow someone to:</p><ul><li><em>Retrieve</em> available <em>slots</em></li><li><em>Create</em> a booking on an available <em>slot</em></li><li><em>Update</em> a booking</li><li><em>Cancel</em> and <em>delete</em> a booking</li></ul><p>Conversational applications will use APIs during dialog management, which is the process that allows an <em>agent</em> or <em>bot</em> to build the response that will be returned to the <em>user</em>.</p><p>First, it collects information from the <em>user</em>’s request and uses that to filter data from the API. The information from the API is then presented to the <em>user</em>, depending on what the response is.</p><p>Continuing with our example, when the <em>user</em> says: “What are the available reservations on Monday morning?”, the conversational application will:</p><ul><li>First, extract the <em>slot</em> <em>bookingPeriod</em> (see <a href="https://medium.com/voice-tech-global/the-difference-between-entities-and-slots-a-definitive-answer-b00a65f5c3dc">our article</a> about <em>slots</em>) “Monday morning”.</li><li>Second, call the reservation API to retrieve all the available <em>slots</em> on Monday.</li><li>Third, search the data available via the API to find what <em>slots</em> match the <em>user</em>’s request.</li><li>Finally, it will present the <em>user</em> with the best options and deliver a response. This may sound something like: “There are two available spots next Monday, one at 8 and the other at 9. Would you be interested in booking one of them?”</li></ul><figure><img alt="A graphic showing how a conversational app extracts data from a user’s request, calls an API to retrieve data, searches the data in an API, and delivers a response." src="https://cdn-images-1.medium.com/max/1024/0*PvOHz1C5dUClzLxB" /><figcaption>Utterance processing and dialog management with API</figcaption></figure><h3>Designing conversation with dynamic data</h3><p>The final step in the process of using an API in conversational AI is presenting the options. This step is also the most challenging.</p><p>The data in an API is dynamic, which means it will vary based on usage. If a <em>user</em> asks “What are the available spots this week?”, one week, the API might return that there is only one opening available, but next week there might be a spot open every day.</p><p>Someone using the app won’t have a reason to know this, and even though the <em>bot</em> can access this information through the API, it’s technically not part of the original request.</p><p>We recommend a specific framework when working with a conversational application that integrates with an API:</p><ol><li>Design the choices</li><li>Design for “no results”</li><li>Design the filter prompts</li></ol><h4>Design the choices</h4><p>This means designing for favourable outcomes, which is when the API returns a limited number of options. This is usually between one and <a href="https://medium.com/google-design/for-conversation-design-three-is-greater-than-seven-54b03d25b101">three</a> choices, since, <a href="https://medium.com/google-design/for-conversation-design-three-is-greater-than-seven-54b03d25b101">as Jonathan Bloom points out</a>, it’s easier for the human brain to remember up to three options.</p><p>The conversational application can then present the choices to the <em>user</em> for a decision.</p><p>For example, when asked for available spots the <em>agent</em> searches the data and finds three openings: Monday 8 to 9, Tuesday 8 to 9, and Friday 9 to 10.</p><p>It can then respond:</p><blockquote>“There are three available reservations this week, Monday at 8 am, Tuesday at 8 am, and Friday at 9 am. Would you like to book any of them?”</blockquote><h4>Design for “no results”</h4><p>When the API returns no results, it is crucial to <strong>always</strong> craft a prompt that offers a solution.</p><p>A response that says: “There are no reservations available this week,” for example, is not helpful.</p><p>Instead, the <em>agent</em> can offer a way for the <em>user</em> to progress in the conversation by saying something like: “All the appointments this week have been booked. <strong>Would you like to check for next week</strong>?”</p><h4>Design the filter prompts</h4><p>When the API returns more than three options that meet the <em>user</em>’s criteria, design a response to investigate approaches. This can be done by asking more questions to the <em>user</em>, thereby further filtering the results.</p><p>These filters need to be specific to the data.</p><p>In our example, each reservation would have a date and a time. If the API returns twenty results, one possible response could be: “<strong>There are a lot of openings this week. Do you have a preferred day?</strong>”</p><p>Following the <em>user</em> response, the application will be able to refine further until it meets the requirements for a favourable outcome. At this stage, we can design the choices.</p><figure><img alt="A graphic showing how to design the responses based on the API results. It highlights the three main rubrics, choices, no results, and filters." src="https://cdn-images-1.medium.com/max/1024/0*agQUx3V50-R30Lwy" /><figcaption>Dynamic data design framework</figcaption></figure><h3>Conversation Design 🤝 Engineering</h3><p>Defining how a bot will respond to dynamic data provides a framework that engineering and dev teams use to build an API integration.</p><p>When it comes to creating documentation and guidelines, Conversation Designers need to contribute, especially when it comes to defining dynamic prompts.</p><p>This is part of the practice of Conversation Design, and is covered in our <a href="https://bit.ly/VTGFCXD">Foundations in Conversational Experience Design (FCxD) course</a> (<strong>next cohort starts October 4th</strong>).</p><p>As part of the course, we review working with stakeholders, team collaboration, and the entire handoff process, in addition to gaining experience with different no-code tools.</p><p>If you liked this story, please give us a clap 👏👏🏻👏🏽.</p><p>Feel free to share in the comments if you have other strategies to design for dynamic data or your own questions about APIs and conversational design.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2980a6dcccd8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/apis-all-you-need-to-know-as-a-conversation-designer-2980a6dcccd8">APIs — All You Need To Know As A Conversation Designer</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Designing Accessible Chatbots: An Introduction]]></title>
            <link>https://medium.com/voice-tech-global/designing-accessible-chatbots-an-introduction-332cead237d4?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/332cead237d4</guid>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[a11y]]></category>
            <category><![CDATA[conversation-design]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[accessibility]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Tue, 07 Sep 2021 23:54:34 GMT</pubDate>
            <atom:updated>2021-09-08T15:19:42.340Z</atom:updated>
            <content:encoded><![CDATA[<p>Imagine coming across the graphic below.</p><figure><img alt="Three small square images that are blurry. (This is intentional.)" src="https://cdn-images-1.medium.com/max/1024/0*CLTlJ6Xu5irbev7v" /><figcaption><em>Which of these three do you prefer?</em></figcaption></figure><p>If I asked: <em>Which of these three do you prefer? </em>Your first thought would probably sound something like:</p><ul><li>“I can’t see what’s in the image.” or</li><li>“Are these images supposed to be so blurry?” or</li><li>“Yikes! I can’t tell what I’m supposed to be looking at!”</li></ul><p>This is the exact same challenge that someone who is visually impaired faces when using a chatbot that relies on visuals to provide a conversational experience.</p><p>In this article, we will discuss how conversation designers can apply the three steps from the <a href="https://www.microsoft.com/design/inclusive/">Inclusive Design Framework</a> to make better chatbot experiences for everyone.</p><figure><img alt="Illustration for the medium post representing a braille table next to the post title with a cursive font" src="https://cdn-images-1.medium.com/max/1024/1*ZwzNNt58svRCk8pVxB6yww.jpeg" /></figure><h3>Step 1: Recognize exclusion</h3><p>To better understand the challenges that visually impaired users face when using a chatbot, we have to step into their shoes. This involves changing the way we think about browsing websites and interacting with chatbots.</p><h4>Try navigating to the chatbot using keyboard only</h4><p><a href="https://www.w3.org/WAI/standards-guidelines/wcag/">Web accessibility guidelines</a> emphasize the support for keyboard compatibility, especially for users who cannot see a mouse pointer on a screen.</p><p>You can test this out for yourself on <a href="https://files.voicetechglobal.com/simulator/accessible-chatbot.html">this website</a>.</p><p>Try to reach the chatbot using the TAB key, then use the enter key to open the bot. For someone who is visually impaired, this is how they would focus on the chat element to use it.</p><p>As a conversation designer, you want to make sure that when chatbot functionality is added to a website, a user can reach it with the keyboard only.</p><h4>Try using a screen reader to test a chatbot</h4><p>This second test involves adding screen reader capabilities to navigate a website.</p><p>As you can guess from the name, screen readers read information about elements on a page, specifically those that a user is ‘on’, or the element that they have TAB-ed to.</p><p>If there is text on the screen, the screen reader will read it out loud for the user. If there is an image, then the screen will read the alt text of that image.</p><p>You can enable screen reader capabilities by:</p><ul><li>Turning on <a href="https://support.apple.com/en-ca/guide/mac-help/mh40578/mac">VoiceOver</a> on a Mac</li><li>Installing <a href="https://www.nvaccess.org/">NVDA</a> on Windows, (please donate if you can to support the maintenance of the application)</li><li>Installing <a href="https://wiki.gnome.org/Projects/Orca">ORCA</a> if you have a Linux distribution</li></ul><p>Once you turn these on, you can use our <a href="https://files.voicetechglobal.com/simulator/accessible-chatbot.html">simulation page</a> again.</p><p>Here’s a bonus test. This time, when you’re navigating using the TAB key, try to dim the light or cover your laptop with a piece of paper. Then try to find the chatbot and have a conversation with it using only your keyboard.</p><p>Navigation and Web accessibility are not something that conversation design can change. But these two tests will help guide you in recognizing how a site can exclude users and make a chatbot conversation better.</p><p>Doing these tests with your own sites can also help you with talking to your webmaster or engineering team if you’re thinking about adding a chatbot to your website.</p><h3>Step 2: Learn from many</h3><p>The tests above can help uncover issues in a product or online platform, but won’t help when it comes to uncovering our own biases.</p><p>This is why it’s important to <em>include</em> the <em>excluded groups of users</em> throughout the process of building a site or product.</p><p>During the research and user testing phases, it is imperative for conversation designers to diversify when it comes to who they interview or ask to test their product. To be specific, you want to include users who are visually impaired in the process so they can evaluate the bot conversation and help identify where there is a <strong>mismatch</strong> with the way they access your chatbot.</p><p>As an example of what you can discover, here are three things we’ve learned from our own user testing:</p><ol><li>Emojis can quickly get annoying as screen readers read their literal descriptions. For example, if your chatbot response is: “we finally launched our website 🚀🚀🚀” with three rockets emojis at the end, the screen reader will read it as: “we finally launched our website rocket rocket rocket.”</li><li>A bot response is always spoken with screen readers. However, there isn’t a way to add nuance or context. For example, abbreviations will be read as is. In the sentence “The Gov. decided to pass the bill.” It’s not clear if “Gov.” means “government” or “governor”.</li><li>Sending multiple small messages can be challenging if there is no obvious numbering that indicates the order of the messages.</li></ol><p>There are many more learnings to uncover from conversations with users or by doing digital ethnography, which is a study of how people react to a real digital environment. There are also many resources available for free, such as this <a href="https://www.facebook.com/groups/1870565113191997">Facebook group</a> for blind and visually impaired users.</p><h3>Step 3: Solve for one and extend to many</h3><p>Steps 1 2 are helpful to uncover where an experience is already broken. Many of the issues uncovered in these tests can be addressed by existing design techniques.</p><p>Let’s go back to our first example of the blurry images.</p><p>The issue with many chatbot conversations is that they are designed to rely on the user looking at the screen.</p><p>To remedy this, we can change the prompt to ask something that the user can answer without having to see an image or visual.</p><p>In other words, we can ask the user for a response that relies on information that can be read.</p><p>Let’s say, for example, we were to provide details and phrase the question: <em>Which would you prefer: hiking in the mud, hiking in the sand, or hiking in the snow?</em></p><p>This allows you to answer the question and complete the experience without needing to see the images or even the screen. It works for those who are visually impaired, <em>and</em>, at the same time, enhances the overall user experience.</p><ul><li>For users that can see the screen, the images supplement their experience and will trigger an emotional response. In our example, with the image only (provided it’s clear), it’s difficult for the user to know how they should respond, such as with a number or general description.</li><li>For users that don’t understand what the images show, rephrasing the question gives them more context.</li></ul><p>We call this technique “<strong>designing for voice only.</strong>”</p><p>The sole principle is to create a conversational experience that can be completed end-to-end with voice, without the screen.</p><p>Working within the constraints of a voice-only design helps ensure that many user groups with different abilities are able to have a great experience of the product. From this perspective, visual elements (images or chatbot buttons) can be added to supplement the experience but are not essential.</p><h3>Why accessible chatbots matter</h3><p><a href="https://ca.linkedin.com/in/daviddame">David Dame</a>, a director of accessibility and mentor in <a href="https://voicetechglobal.com/foundations-in-conversational-experience-design/">our course program</a>, says it best:</p><blockquote><strong>I have cerebral palsy, but my money doesn’t. If you want me to buy your product then design one that I can use. If it’s created with </strong><a href="https://www.a11yproject.com/posts/2017-08-26-a11y-and-other-numeronyms/"><strong>a11y</strong></a><strong> in mind, I will show you the money.</strong></blockquote><p>Resolving accessibility issues is often seen as a good deed, but in reality, designing for people with disabilities is designing for all of us. Making the adjustments to the design and build process can lead your conversational products to a larger audience that includes everyone, not just those who don’t identify as being visually impaired.</p><p><a href="https://medium.com/@gtonye/5-tests-to-audit-and-improve-the-accessibility-of-your-chatbot-ec0298c43e8c">This guide</a> contains more techniques to help you identify broken experiences and implement technical and design solutions to make your product more accessible.</p><p>Inclusive design covers accessibility and more. In our Advanced Conversational Experience Design Course, we dedicate an entire week to this topic. <a href="https://voicetechglobal.com/advanced-conversational-experience-design-course/">Find the full syllabus here</a>.</p><p>If you liked this story, please give us one of Medium’s <em>clapping hands clapping hands light skin tone clapping hands medium skin tone</em>.</p><p>We are always looking for examples of how we can make conversation design more accessible. Feel free to share in the comments if you have built or know of an accessible chatbot in real life.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=332cead237d4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/designing-accessible-chatbots-an-introduction-332cead237d4">Designing Accessible Chatbots: An Introduction</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5 Tests To Audit And Improve The Accessibility Of Your Chatbot]]></title>
            <link>https://medium.com/voice-tech-global/5-tests-to-audit-and-improve-the-accessibility-of-your-chatbot-ec0298c43e8c?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec0298c43e8c</guid>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[conversation-design]]></category>
            <category><![CDATA[inclusive-design]]></category>
            <category><![CDATA[a11y]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Tue, 07 Sep 2021 23:53:41 GMT</pubDate>
            <atom:updated>2021-09-07T23:59:31.379Z</atom:updated>
            <content:encoded><![CDATA[<p>With a growing number of businesses and organizations using Chatbots on their website, we wanted to provide a framework to make the Conversational AI interface accessible and inclusive.</p><p>In <a href="https://voicetechglobal.com/conversation-design-training/">our courses</a> and <a href="https://voicetechglobal.com/workshops/">workshops</a>, we focus on outlining practical steps and techniques to recognize where the experience is broken then design solutions to make it more exclusive.</p><p>The five tests guide will help Conversation Designers:</p><ul><li>Audit a Chabot</li><li>Suggest actionable solutions with the engineering team or webmaster</li><li>Enhance their design to make it more accessible</li></ul><figure><img alt="Illustration for the post with the icon of a checklist and the title of the post with a cursive font" src="https://cdn-images-1.medium.com/max/1024/1*5Rr7smFIP1-I8_Hp9BhNWQ.jpeg" /></figure><h3>Test 1: You can reach the chatbot with your keyboard only</h3><h4>How to test</h4><p>After going on the page without your mouse, try to get to the chatbot using the keyboard only. The standard approach is to use the tab key. The TAB key will help you navigate between the elements of the page. You can use the SHIFT and TAB keys to navigate back to the previous element.</p><p>The first thing you can evaluate is when you move with the TAB key: do the elements show a focus when “tabbed” to?</p><p>The focus looks like an overlay rectangle around the element. The rectangle only has borders that seem to surround the component.</p><figure><img alt="Example of a focused button with a blue rectangle overlay surrounding the element" src="https://cdn-images-1.medium.com/max/228/0*DIvG6cmKxQnddSp_" /><figcaption>Example of a focused button</figcaption></figure><p>If and when you reach the chatbot icon, the second thing you want to assess is if you can open the chatbot by pressing the RETURN (or ENTER) key.</p><h4>Why is it a problem</h4><p>For a user leveraging assistive technology, they will never be able to reach the chatbot.</p><h4>Recommendations</h4><p>The developer of the website can solve this by making sure that:</p><ul><li>Option 1: the Chabot icon is a <em>&lt;button&gt;</em> element in the HTML</li><li>Option 2: the chatbot element (if it’s not a <em>&lt;button&gt;</em>) has a <a href="https://developer.mozilla.org/en-US/docs/Web/HTML/Global_attributes/tabindex"><em>tabindex</em></a> attribute that will make it “focusable” and allow users to TAB to it.</li></ul><h3>Test 2: After opening the chatbot you can immediately type</h3><h4>How to test</h4><p>When you click (or TAB then RETURN) on the chat icon, verify that it immediately focuses on the place where you can type the text.</p><p>The focus on the place where you type will have the blinking cursor on the section where the user is expected to type. It can also have the rectangle overlay surrounding the input section.</p><figure><img alt="Example of focus on the text input with the cursor appearing over the place holder “Enter the bot response”" src="https://cdn-images-1.medium.com/max/1024/0*M3zOgBZCqMYIPiTk" /><figcaption>Example of focus on the text input</figcaption></figure><h4>Why is it a problem</h4><p>As you have experienced during our test 1, TAB-ing to an element is quite difficult, especially when the page has a lot of elements. After clicking on the Chatbot icon, if the focus goes to another place in the page, the user will have to TAB their way back to that chatbot input to enter the text to speak with the bot.</p><h4>Recommendations</h4><p>The developer of the website can enforce the focus on the input using Javascript.</p><p>See guidelines and examples on<a href="https://developer.mozilla.org/en-US/docs/Web/API/HTMLElement/focus"> the Mozilla documentation</a>.</p><h3>Test 3: Any prompt can be answered without looking at the screen</h3><h4>How to test</h4><p>When interacting with the bot, you want to ensure that you are able to converse without needing to rely on the graphical elements. If the bot responds with images or buttons (suggested actions), you can ask yourself the following questions: “If I did not have access to the images or the, would I still be able to engage with the chatbot?<em>”</em></p><p>For example, if the bot says, “Hi, Guest! I love interacting with real people. Choose an option below so we can start.” If the user isn’t able to see or access the options in the buttons, the prompt isn’t helpful for the user to progress in the conversation.</p><figure><img alt="Example of the prompt from the bot relying on the screen and buttons" src="https://cdn-images-1.medium.com/max/688/0*Ps0B98_ewonaDC2r" /><figcaption>Example of the prompt from the bot relying on the screen and buttons</figcaption></figure><figure><img alt="Example of the prompt from the bot relying on the screen with the buttons hidden" src="https://cdn-images-1.medium.com/max/688/0*9eMFr1zBDFsPKHcv" /><figcaption>Example of the prompt from the bot relying on the screen with the buttons hidden</figcaption></figure><h4>Why is it a problem?</h4><p>As described in the first test, a user would need to TAB through all the options to carry on that conversation.</p><h4>Recommendations</h4><p>Making the prompt actionable, with a more concise version of the options, can help the user quickly understand what to do without needing to navigate through the buttons.</p><p>The technique of making sure that the prompt is self-sufficient is increasingly adopted in multimodal design for conversational products.</p><h3>Test 4: All images have an alternate text</h3><h4>How to test this?</h4><p>When using images in the chat, it’s important to add <em>alt</em> text. Otherwise, the screen reader will say “missing description” or “unlabeled image.”</p><p>To verify if an image is screen-reader friendly you can:</p><ul><li>Use a screen-reader, or Accessibility equivalents (on <a href="https://support.apple.com/en-ca/guide/mac-help/mh40578/mac">Mac</a>, <a href="https://www.nvaccess.org/download/">Windows</a>, and <a href="https://wiki.gnome.org/Projects/Orca">Linux</a>), navigate to the image, and hear what the screen-reader software says</li><li>Right-click on the image and use the web developer tools by clicking on “Inspect” in the contextual menu. Then in the code showing in the developer tools, search for the <em>img</em> tag and an attribute alt.</li></ul><figure><img alt="Image of the contextual menu and the Inspect highlighted with a pink rectangle" src="https://cdn-images-1.medium.com/max/1024/0*lo0gwz09Vob0oPKt" /><figcaption>Image of the contextual menu and the Inspect</figcaption></figure><figure><img alt="Image of the developer tools after clicking on Inspect with a highlight on the image tag." src="https://cdn-images-1.medium.com/max/1024/0*G7LOFiYiK-VnRnVR" /><figcaption>Image of the developer tools after clicking on Inspect</figcaption></figure><p>In the snippet below, you are looking for any indication of an alt attribute in the tag</p><blockquote>&lt;img src=”https://s3.amazonaws.com/com.getstoryflow.api.images/1621532012819-2.png&quot;&gt;</blockquote><p>There is none, but if there was it would look something like:</p><blockquote>&lt;img src=”https://s3.amazonaws.com/com.getstoryflow.api.images/1621532012819-2.png&quot; <strong>alt=”blurry image showing a phone”</strong>&gt;</blockquote><h4>Why is it a problem?</h4><p>If you have done the test with a screen-reader solution, you must have experienced how degraded the experience can be when the image has no description.</p><p>If you have not used a screen-reader, the reader will say “missing description” or “unlabeled image” and for the user, they are missing out on any information that the image was communicating.</p><h4>Recommendations</h4><p>The solution to this problem involves the conversation designer AND the developer. On the developer end, it’s important to ensure that all images in the conversation have an alt tag.</p><p>For the conversation designer:</p><ul><li>It’s important to ensure that the images augment the experience but are not essential for it. Similar to test 3, the user should be able to carry on the conversation without the image as much as possible.</li><li>It’s crucial to supply the alternate description for any images they use.</li></ul><h3>Test 5: Screen-reader by friendly design</h3><h4>How to test this</h4><p>If you are using a screen-reader software (<a href="https://support.apple.com/en-ca/guide/mac-help/mh40578/mac#:~:text=To%20change%20these%20preferences%2C%20choose,click%20Accessibility%2C%20then%20click%20VoiceOver.&amp;text=Turn%20VoiceOver%20on%20or%20off.&amp;text=Learn%20how%20to%20use%20VoiceOver.&amp;text=Customize%20VoiceOver%20settings.">VoiceOver</a> on Mac, <a href="https://www.nvaccess.org/download/">NVDA</a> on Windows, and <a href="https://wiki.gnome.org/Projects/Orca">Orca</a> on Linux), you want to browse the history of the conversation and verify that it is readable. If you listen to the conversation TAB through, it reads each utterance one after the other, the experience is not ideal for the user.</p><p>If you are not using a screen reader, you can again use the developer tools inspection by right-clicking on any text, the bot or yours and click inspect.</p><p>In the code you are looking for either:</p><ul><li>An aria-label tag on the text, specifying “the bot said” or “the user said”</li><li>Find anywhere in the text a snippet saying “the bot said” or “the user said” with an element with the class <em>sr-only</em></li></ul><figure><img alt="Example of element with the sr-only class highlighted with a purple rectangle inside the developer tools" src="https://cdn-images-1.medium.com/max/1024/0*2vIrj2TIMRyyGiwT" /><figcaption>Example of an element with the sr-only class</figcaption></figure><p>Or a snippet of code looking like this</p><blockquote>&lt;section tabindex=”0&quot; class=”bot message-bubble” <strong>arial-label=”at 21:4 the bot said”</strong>&gt;Ok, Pick up , got it. Do you want that pizza right now, or would you like it at a later time?&lt;/section&gt;</blockquote><h4>Why is it a problem?</h4><p>If I showed this transcript for you:</p><ul><li>Hi, how are you</li><li>Good thank you</li><li>What is your ID?</li><li>How can I check my balance?</li><li>One moment please</li></ul><p>Would you be able to find who is the user and who is the bot?</p><p>That is the challenge that a user with assistive technology listening to the conversation history faces.</p><h4>Recommendation</h4><p>The developer for the website should add an sr-only class. <a href="https://gist.github.com/ffoodd/000b59f431e3e64e4ce1a24d5bb36034">This resource</a> will provide guidance.</p><p>Using that sr-only class, it will allow inserting screen reader only text that will make the conversation more meaningful for users that don’t have access to all the visual cues of the chatbot.</p><h3>How to use the 5 Tests guide?</h3><p>The 5 tests guide can be used on any existing or soon-to-be-published chatbots. We recommend sharing out the guide internally to kickstart the conversation about accessibility for the chatbot.</p><p>The 5 tests and their recommendation are a small part of our techniques for Inclusive Design in Conversational AI. If you want to learn more about our Inclusive Conversational Design Toolkit, get in touch with us at hello@voicetechglobal.com.</p><p>If you liked this story, please give us a clap 👏👏🏻👏🏽.</p><p>Feel free to share in the comments If you have suggestions or additional tests that we could include to continue to help make the Chatbot experience more accessible.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec0298c43e8c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/5-tests-to-audit-and-improve-the-accessibility-of-your-chatbot-ec0298c43e8c">5 Tests To Audit And Improve The Accessibility Of Your Chatbot</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Difference Between Entities and Slots: A Definitive Answer]]></title>
            <link>https://medium.com/voice-tech-global/the-difference-between-entities-and-slots-a-definitive-answer-b00a65f5c3dc?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/b00a65f5c3dc</guid>
            <category><![CDATA[conversation-design]]></category>
            <category><![CDATA[named-entity-recognition]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Fri, 03 Sep 2021 00:57:50 GMT</pubDate>
            <atom:updated>2021-09-09T21:58:18.202Z</atom:updated>
            <content:encoded><![CDATA[<p>Here is a quick test for you. Take a look at this phrase from a person using a bot to gather information:</p><p>“<em>Tell me about the training to become a Conversation Designer.</em>”</p><p>Which of the following would you say is correct?</p><ol><li>The <em>intent</em> is <em>TrainingInquiry</em> and “Conversation Designer” is the <strong>Value</strong> of the <em>slot</em> “Profession”.</li><li>The <em>intent</em> is <em>TrainingInquiry</em>, “Conversation Designer” is the <strong>Slot</strong>.</li><li>The <em>intent</em> is <em>TrainingInquiry</em>, “Conversation Designer” is the <strong>Entity</strong>.</li></ol><p>If you have no idea what the answer is, you’re not alone. When we ask this question in our <a href="https://voicetechglobal.com/conversation-design-training/">Conversation Design (CxD) Training classes</a>, we get all three answers.</p><p>In this article, we will review where the confusion comes from and what CxD looks like using real-world examples. By the end, you will know how to ace this quiz.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*amFnT5f3qndkKuhBv46y0w.jpeg" /></figure><h3>What happens when a bot processes what you say</h3><p>Let’s start with unpacking what happens when a <em>user</em> interacts with a <em>bot</em>.</p><p>Take a Space Agency Bot that is capable of helping someone book a ticket on the next rocket ship going into space.</p><p>The <em>user</em> says “I want to book a ticket to go to Mars in 2023,” and a number of things happen in seconds.</p><p><strong>First</strong>, the machine learning system (or ML, for short) in the <em>bot</em> tries to identify if that sentence matches one of the <em>intents</em> it has been trained on. An <em>intent</em> is a goal that the <em>user</em> has that is programmed into the <em>bot</em>. Usually, the ML system is programmed with more than one <em>intent</em>.</p><p>For the Space Agency <em>bot</em> in this case, it has been programmed with an <em>intent</em> called <em>BookTicket</em>. Other potential <em>intents</em> could be <em>CancelTicket</em>, or <em>ChangeBooking</em>.</p><p><strong>Second</strong>, to book the ticket, the <em>bot</em> needs the destination, the trip date, and the number of desired tickets. This is required information for an <em>intent</em>. The technical term for this type of information is <em>slots</em>.</p><p>The Space Agency bot will try to fill<em> in</em> the <em>slots</em> using the information in the <em>user</em> utterance (or what the <em>user</em> asked for). “Mars”, “2023” and “one” (deducted from the “a”) are good candidates to fill the <em>slots</em>. These details are called <em>slot values</em>.</p><p><strong>Finally</strong>, the <em>bot</em> provides a summary result. Here are the results from the Space Agency example:</p><p><em>Intent</em>: <em>BookTicket</em></p><p><em>Slots</em>:</p><ul><li>“Destination” with a <em>Slot Value</em> of “Mars”</li><li>“Trip Date” with a <em>Slot Value </em>of “2023”</li><li>“Number of tickets” with a <em>Slot Value </em>of “one“</li></ul><figure><img alt="The image describes the steps that happen when a bot processes a user utterance" src="https://cdn-images-1.medium.com/max/1024/1*0qoQ0I_UdxqhLkfz7TWZEA.png" /><figcaption>User Utterance Processing By ML</figcaption></figure><h3>How Conversation Designers train bots to understand humans</h3><p>To get a <em>bot</em> to understand a request, it needs training.</p><p>The training consists of feeding the ML system that the bot uses with <em>intents</em>.</p><p>For all the <em>intents</em> the conversation designer wants a <em>bot</em> to detect, they provide a name and a list of annotated phrases.</p><p>The conversation designer for the Space Agency bot names the book a ticket <em>intent</em>: <em>BookTicket</em>.</p><p>Here are a few annotated phrases they might provide:</p><ul><li>“I want to get {NumberOfTicket} {TripDate} ticket to {Destination}”</li><li>“I want to book {NumberOfTicket} ticket to {Destination} in {TripDate}”</li></ul><p>{NumberOfTicket}, {TripDate}, and {Destination} are the <em>slots</em> we mentioned earlier. They represent the information necessary for conversation designers to include so that the <em>bot</em> can return an accurate response to the <em>user</em>, as long as it has detected the right <em>intent</em>.</p><h3>Understanding Slot Types</h3><p><em>Slot types</em> are crucial because they indicate the possible values that can be in a <em>slot</em>. For the machine to train, it will need to know what the types of <em>slots</em> are.</p><p>There are a few standard <em>slot types</em>. These are:</p><ul><li>Text: Any word or group of words.</li><li>Numeric: Any number, “1” or “one”.</li><li>Date: Any expression of a date, for example, “Tomorrow”, “March 2042”.</li><li>Time: Any expression of time, for example, “Four forty-four PM”, “Sixteen fourty-four” or “half past nine”.</li></ul><p>In our example, the {NumberOfTicket} will be of the Numeric slot type, and {TripDate} will be of the Date slot type.</p><p>But what will be the <em>slot type</em> of {Destination}? We can use Text, but since this <em>slot type</em> allows any words, it will not be specific enough. A user could say they want to go ‘there’ and would get nowhere.</p><p>To solve that, when training the ML system, conversation designers can create something called a <em>custom</em> <em>slot type</em>. These <em>slot types</em> need a name, a list of values, and possible synonyms for each value.</p><p><strong><em>Custom slot types</em> are also called <em>entities</em></strong>. An <em>entity</em> is a finite list of <em>values</em> and synonyms that can be used as a <em>slot type</em> when none of the standard types are a good fit (ie. text, numeric, date, and/or time).</p><p>In our Space Agency bot example, we can define an <em>entity</em> with the name<em> </em><strong><em>CelestialObject</em></strong>, and give it many values, for example:</p><ul><li>“Mars”<em> </em>with the synonyms<em> </em>“red planet”, “4th planet”</li><li>“Moon”</li></ul><figure><img alt="The image describes the steps and the assets (intent, slot, slot type, entities) required for ML training" src="https://cdn-images-1.medium.com/max/1024/1*GcZ_Ky3JMQutORllw5LuSg.png" /><figcaption>ML Training Process</figcaption></figure><p>When we assemble everything for the Space Agency Bot, we feed ML with:</p><p><em>Intent</em>: <em>BookTicket</em></p><p><em>Phrases</em>:</p><ul><li>“I want to get {NumberOfTicket} {TripDate} ticket to {Destination}”</li><li>“I want to book {NumberOfTicket} ticket to {Destination} in {TripDate}”</li></ul><p><em>Slots</em>:</p><ul><li>NumberOfTicket (Type: Numeric)</li><li>TripDate (Type: Date)</li><li>Destination (Type: Entity, CelestialObject)</li></ul><h3>Where does the confusion between Entities and Slots come from?</h3><p>From the last section we have the following:</p><ul><li><em>Slot</em>: a required piece of information for the <em>intent</em>.</li><li><em>Slot Type</em>: the kind of information that a <em>slot</em> can have.</li><li><em>Entity</em>: a <em>custom type</em> for a <em>slot</em> with a finite list of <em>values</em>.</li></ul><p>When we look at all the available conversation design tools, the terminology above tends to get a little confusing. As you discover more, you’ll learn that different platforms use different terminologies for each concept:</p><ul><li>Amazon, for its Alexa Development, uses <em>slots</em>, <em>slot types</em> and <em>slot values</em>. Amazon defines <em>entities</em> as <em>custom slot types</em>.</li><li>Google, in Dialogflow, solely uses <em>entity</em>. <em>Slot types</em> are <em>entity types</em>, <em>slot values</em> are <em>entity</em> entries. The one interesting thing is, the terminology Google uses changes again. In the next step after <em>intent</em> detection, <a href="https://cloud.google.com/dialogflow/es/docs/fulfillment-webhook-slot-filling">Google uses the term… “slot” filling</a>.</li><li>IBM Watson is similar to Google in using solely <em>entity</em> with <em>entity name</em> for <em>slot type</em>.</li><li>Voiceflow uses <em>entity</em> and <em>entity type</em> instead of <em>slot</em> and <em>slot types</em>.</li><li>Botmock uses <em>variable</em> and <em>variable types</em> for <em>slot</em> and <em>slot types</em>, and also uses <em>entity</em> as the custom <em>slot type</em>.</li></ul><figure><img alt="The image describes the terminologies use by each platform for intent, slot, slot type and entity." src="https://cdn-images-1.medium.com/max/1024/1*Bvpey25UbfE79Z3G2q_e5Q.png" /><figcaption>Terminologies Broken By Platform</figcaption></figure><p>Let’s look back at our earlier quiz.</p><p><em>“Tell me about the training to become a Conversation Designer.”</em></p><p>Which of the following would you <strong>now</strong> say is correct about this <em>user</em> utterance?</p><ol><li>The <em>intent</em> is <em>TrainingInquiry</em> and “Conversation Designer” is the <strong>value</strong> of the <em>slot</em> “Profession”.</li><li>The <em>intent</em> is <em>TrainingInquiry</em>, “Conversation Designer” is the <strong>slot</strong>.</li><li>The <em>intent</em> is <em>TrainingInquiry</em>, “Conversation Designer” is the <strong>entity</strong>.</li></ol><p>The answer is … <strong>1</strong> 🙂.</p><p>Remember: <em>Slot</em> and <em>entity</em> are the terms used to describe the information that conversation designers feed into machines. A <em>slot</em> can take the following<em> slot types</em>: dates, times, text, and numbers. <em>Entities</em> are specific to custom <em>slot types</em>, like a ‘planet’ custom slot type, for example.</p><p>In classes at <a href="https://voicetechglobal.com/">Voice Tech Global</a>, we recommend our students focus on the definition of <em>slot</em> and <em>entity</em> when working with the technologies available. Then they can navigate between tools and platforms without confusion because they understand how terms are defined first.</p><h3>Test yourself</h3><p>Now that you’re a <em>slot</em>, <em>intent</em> expert, it’s time to do a little test.</p><p>Take <a href="https://forms.gle/sW3NSMpUzCgYHtGJ6">this quiz</a> to verify your understanding of the concepts. 🙂</p><p>P.S.: There is a surprise at the end if you get everything right!</p><h3>What does CxD training actually look like?</h3><p>Once conversation designers have their <em>intent</em> definitions, with the <em>slots</em> and <em>entities</em>, the actual machine training begins.</p><p>In our <a href="https://bit.ly/VTGFCXD">Foundations in Conversational Experience Design (FCxD) course</a>, we cover the entire process in different no-code tools. We focus on helping those interested in conversational design begin a career as a practitioner.</p><p>Students who take the FCxD course leave with real-world experience, a polished portfolio, and are introduced to a whole new world of employment opportunities. <a href="https://bit.ly/VTGFCXDSyllabus">Find the full syllabus and course details here.</a></p><p>If you liked this story, please give us a clap 👏👏🏻👏🏽.</p><p>Feel free to share in the comments if our guide has helped you understand how bots work, the difference between <em>slot</em> and <em>entity</em>, or your own questions about conversational design.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b00a65f5c3dc" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/the-difference-between-entities-and-slots-a-definitive-answer-b00a65f5c3dc">The Difference Between Entities and Slots: A Definitive Answer</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Machine Learning Confidence Scores — All You Need to Know as a Conversation Designer]]></title>
            <link>https://medium.com/voice-tech-global/machine-learning-confidence-scores-all-you-need-to-know-as-a-conversation-designer-8babd39caae7?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/8babd39caae7</guid>
            <category><![CDATA[conversation-design]]></category>
            <category><![CDATA[conversational-interfaces]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[dialogflow]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Wed, 25 Aug 2021 13:32:46 GMT</pubDate>
            <atom:updated>2021-09-04T14:48:18.654Z</atom:updated>
            <content:encoded><![CDATA[<h3>Machine Learning Confidence Scores — All You Need to Know as a Conversation Designer</h3><p>Recently we advised a government organization on how to understand why their Interactive Voice Response (IVR) agent failed on a particular user request.</p><p>This organization has a self-serve conversational application that allows constituents to manage and renew their driving licenses, vehicle tabs, and professional licenses. However, when users stated, “I want to renew my license,” the bot would always surface a message indicating that it did not understand the utterance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*__bZcWNGta5KV7mJnSS-JQ.png" /><figcaption>Transcripts of three conversations with the Gotham IVR</figcaption></figure><p>We had two puzzling questions here:</p><ul><li>To which Intent (“Driving License,” “Vehicle Tab,” or “Professional License”) do you add the utterance?</li><li>Why does the system return a No Match despite having “renew my driving license,” “renew my license plate,” and “renew my real estate license” in the utterances of our language model?</li></ul><p>In this article, we will walk you through how we solved this puzzle thanks to our understanding of Machine Learning confidence scores. We’ll also touch on how they impact the conversational products we are building as Conversation Designers and how we can mitigate their impact.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XEixruND7F75o-xgnWicaw.jpeg" /></figure><h3>What is a Confidence Score?</h3><p>A Confidence Score is a number between 0 and 1 that represents the likelihood that the output of a Machine Learning model is correct and will satisfy a user’s request.</p><p>The output of all Machine Learning (ML) systems is composed of one or multiple predictions. For example, YouTube ML will predict which video(s) you want to see next; Uber ML will predict the ETA (estimated time of arrival) for a ride.</p><p>Each prediction has a Confidence Score. The higher the score, the more confident the ML is that the prediction will satisfy the user’s request.</p><p><a href="https://docs.microsoft.com/en-us/azure/cognitive-services/qnamaker/concepts/confidence-score">Microsoft’s breakdown</a> of Confidence Score’s meaning for Conversational AI can be categorized as follows:</p><ul><li>Over 0.7: the prediction is a strong candidate for answering the user query.</li><li>Between 0.3 and 0.7: the prediction can partially answer the request.</li><li>Below 0.3: the prediction is probably not a good choice.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dsbBOHQE9hqDOVSNL06qFQ.png" /><figcaption>Mapping between confidence scores and their meaning</figcaption></figure><p>Confidence Scores are the result of Machine Learning model training.</p><h3>Confidence Scores in Conversational AI</h3><p>In Conversational AI, ML is essential in many stages of the processing of the user request:</p><ul><li>During Natural Language Understanding (NLU): ML helps predict the Intent (which represents what the user is looking for) from an utterance (what the user said or typed).</li><li>During Automated Speech Recognition (ASR): ML will predict the transcription from the audio of what the user said.</li><li>During Sentiment or Emotion Analysis: ML predicts the sentiment (generally as positive, negative, or neutral) or the emotion based on the user utterance or the conversation (back and forth between the user and the agent) transcript.</li><li>During Natural Language Generation (NLG): ML will predict what to answer from the user utterance.</li><li>During Text-To-Speech (TTS): ML will predict the audio from the answer text in NLG.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xKBEk5-OVA0oL1P3UKKDbQ.png" /><figcaption>Machine Learning impact on Conversational AI</figcaption></figure><p>As mentioned above, each prediction will have a confidence score.</p><p>One consequence of this heavy use of ML is that for Voice applications, the predicted Intent combines the confidence scores of the ASR and the NLU stages. The compounded risk of misinterpretation presents a great risk for Conversation Designers when they assume that the predictions are always correct.</p><h3>Why Confidence Scores Matter For Conversation Designers</h3><p>In our government organization case study, when the user said, “I want to renew my license,” the bot responded, “I’m sorry, we can’t help with that. Are you looking for information about driving licenses, vehicle tabs, or professional licenses?”</p><p>The first reaction is often to blame the lack of training data and add the utterance in one of our three intents. Because our three Intents — Driving License, Vehicle Tab, and Professional License — can all have that utterance, that option was not available to us.</p><p>No Match was suspicious because even if ML doesn’t do keyword matching, renewing the license was very close to some of the existing utterances in the language model. For example, “driving license renewal,” “renew my architect license,” or “renew my license plate.”</p><p>We decided to investigate what the system understood.</p><p>The team used <a href="https://cloud.google.com/dialogflow">Google Dialogflow</a> as the NLU engine with the <a href="https://cloud.google.com/dialogflow/es/docs/integrations/phone-gateway">Phone Gateway</a> for the channel on this project. One key advantage for debugging is that in the Dialogflow console, you can visualize the breakdown of the API results when testing, which shows all the details of the Machine Learning results.</p><p>We realized that when the user said “renew my license,” the ML predicted a No Match with a 0.8 score. This led us to discover that in the Agent Settings, there is a parameter called <em>classification threshold</em> (under the ML section).</p><figure><img alt="The image shows the Machine Learning settings in Dialogflow CX" src="https://cdn-images-1.medium.com/max/1024/1*q4zFzkxpnK4UjR3N39M51g.png" /><figcaption>Machine Learning settings in Dialogflow</figcaption></figure><p><a href="https://cloud.google.com/dialogflow/es/docs/intents-matching#confidence">This documentation</a> states that the machine predicts multiple intents. It will return a No Match if the Intent with the highest confidence score is below the threshold.</p><p>We finally figured out that the ML was actually matching one of the Intents, but the threshold was hiding it. We discovered this by lowering the confidence score threshold.</p><figure><img alt="The image shows the JSON response from the natural language engine and highlights the 0.4 confidence score for the professional license intent" src="https://cdn-images-1.medium.com/max/1024/1*TlYU_Mi4isXEDrkGly3sBg.png" /><figcaption>Diagnostic info in Dialogflow CX simulator</figcaption></figure><p>To provide a better response for a No Match when the Confidence Score met the threshold (e.g., a Confidence Score and threshold both of 0.8), we added in additional logic to change the response for that question.</p><p>Now, instead of “<em>I’m sorry, we can’t help with that. Are you looking for information about driving licenses, vehicle tabs, or professional licenses?</em>” the agent responds with “<em>Driving license, vehicle tabs or professional?</em>”.</p><blockquote>We often hear that “<em>chatbots don’t work</em>” or “<em>I asked Alexa, and it didn’t know</em>.”</blockquote><p>The standard approach to resolve these issues is to add more training data to the model. But as we discovered, it won’t fix the issue if the problem is happening at the Machine Learning level.</p><p>We now know that a No Match response has the following causes:</p><ul><li><strong>ML did not predict the Intent</strong>: In this situation, you can test adding more training data.</li><li><strong>ML predicted the Intent</strong>, but some underlying rule prevents the matched Intent from being picked up: You need to dig into the tool or platform settings to determine if a rule is causing an unexpected issue.</li></ul><h3>How Conversation Designers Can Access Confidence Scores</h3><p>You may be wondering if the process we used is something a Conversation Designer can do? The answer is “yes.”</p><p>Below we show a (non-exhaustive) list of standard tools/platforms and how you can access information about Confidence Scores.</p><h3>Alexa Developer Console</h3><p>Alexa does not show the actual Confidence Score but provides two useful data points to debug the NLU issues.</p><ul><li>The first one is the <a href="https://developer.amazon.com/en-US/docs/alexa/custom-skills/test-utterances-and-improve-your-interaction-model.html">utterance profiler</a>: When you look at your Intents in the console, you can press “Evaluate Model” to access it.</li><li>The second one is the <a href="https://developer.amazon.com/en-US/docs/alexa/devconsole/test-your-skill.html#intent-debugging">Device Log</a>: If you are building a custom skill, the device log shows you Considered Intents. It’s helpful to find competing intents and understand why the application behaves a certain way.</li></ul><figure><img alt="the image shows the Alexa Developer Console and the device log information with a highlight on the other considered intent" src="https://cdn-images-1.medium.com/max/1024/1*6Az1vphswEQFtQxRDGAwGA.png" /><figcaption>Device log in the Alexa Developer Console simulator</figcaption></figure><h3>Google Dialogflow</h3><p>In the simulator, you can view the Diagnostic information as you test, which will show you the Confidence Score. The downside is that it only shows the intent with the highest confidence score and not the other considered Intents.</p><p>We would recommend for Conversation Designers to know that:</p><ul><li><a href="https://cloud.google.com/dialogflow/es/docs/agents-settings#ml">The classification threshold</a> is essential as it impacts intent detection and, therefore, the way you design your conversation.</li><li><a href="https://cloud.google.com/dialogflow/es/docs/knowledge-connectors#settings">The knowledge results preference</a> orchestrates the balance between knowledge base results and the results from the Intents.</li></ul><h3>Rasa X</h3><p>Rasa allows full control over the Conversational AI process. One of the great features is the ability to <a href="https://rasa.com/docs/rasa-x/user-guide/review-conversations/">review conversations</a>.</p><p>The interface provides a complete breakdown of all Confidence Scores and ML information during the conversations between users and the bot.</p><figure><img alt="The image shows the Rasa X interface for conversations explorer with a highlight on the ability to view the confidence scores and the other considered intents with their scores as well" src="https://cdn-images-1.medium.com/max/1024/1*PVxtfJPHSQc5-MqVnPHMfA.png" /><figcaption>Reviewing conversations in Rasa X</figcaption></figure><h3>All the other tools</h3><p>We can’t list all the available setups out there, but for the tool you’re using, here are few things that you can do:</p><ul><li>Reach out to the tool developer and ask them if they have any parameters for Confidence Scores that you should be aware of.</li><li>Ask the developers on your team if they know or can investigate.</li><li>Reach out to the <a href="https://join.slack.com/t/voicetechglobal/shared_invite/zt-3m0lljif-Cl9zsh~NoOj5ENtQ0OF1og">Voice Tech Global Slack Community</a> :)</li></ul><h3>Using Confidence Scores In CxD</h3><p>This article covers Confidence Scores for problem-finding purposes only, but you can also use the knowledge of Confidence Scores for disambiguation in Conversation Design.</p><p>If you want to know more, you can join our <a href="https://voicetechglobal.com/advanced-conversational-experience-design-course/">ACXD Course</a> to learn about our strategies for designing for and with ambiguity in mind.</p><p>If you liked that story, please give us a few claps 👏.</p><p>We would love to hear your own experience with confidence scores and language model optimization as Conversation Designers in the comments!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8babd39caae7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/machine-learning-confidence-scores-all-you-need-to-know-as-a-conversation-designer-8babd39caae7">Machine Learning Confidence Scores — All You Need to Know as a Conversation Designer</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Product Builders Can Mitigate Bias In Conversational AI]]></title>
            <link>https://medium.com/voice-tech-global/how-product-builders-can-mitigate-bias-in-conversational-ai-2012d822da95?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/2012d822da95</guid>
            <category><![CDATA[bias]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[conversational-ai]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Thu, 02 Jul 2020 14:20:06 GMT</pubDate>
            <atom:updated>2020-07-02T14:20:06.176Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8klooBkUs_I_LV2UBRvzA.png" /></figure><h3>Conversational AI is about assembling building blocks</h3><p>Conversational AI is about assembling building blocks and for me, it was my first exposure to Artificial Intelligence. After a couple of years of interacting with it, I quickly fell in love with using it. What I liked most about conversational AI was how accessible it is to get started and do something with it.</p><p>Building conversational products today is more about putting pre-built or pre-trained pieces together than becoming a machine learning expert. When I have to describe it to anyone, it looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jCj8x5QVJgFbJTWAHkuaBg.png" /><figcaption>An example of a weather conversational application</figcaption></figure><p>In the example above, if you build an Alexa skill or an action on Google, you are not in charge of implementing the speech-to-text, the Google Nest Hub or the smart speaker handle it, or the actual natural language understanding (i.e., the algorithm that will take a text and match it to a predefined understanding). You can use building blocks that are already made available by Amazon, Google, Microsoft, or IBM.</p><h3>AI is a reflection of its creators and their biases</h3><p>In light of recent events that occurred in the United States, highlighting the problems of systemic racism, and the protests that followed around the world, we have been looking at ways to address the struggles faced by the black community.</p><p>In these conversations, technology has not been scrutinized as much, mostly because these systems, rooted in mathematics, are perceived to be based on logic, not human interaction. But when we reflect deeper on it, our existing technology has a single flaw that no proof or algorithm can easily solve: <strong>It is built by humans</strong>. Therefore, human failings can be built into the machine, however unaware we are of them.</p><p>I decided to reflect on conversational AI. As seen in the diagram above, when we build our conversational AI, we rely on that block from Google or Amazon to complete the speech to text functionality. If we look at <a href="https://www.pnas.org/content/pnas/117/14/7684.full.pdf">this recent Stanford study</a>, a lot of people may never hear from your conversational agent because when the device does not do a proper speech to text, it never goes into the next block and therefore African Americans will face difficulties accessing your services.</p><p>As I started digging deeper, I realized that the systems we rely on in these products, convenient APIs (for sentiment analysis for example) or pre-trained language models were prone to something called bias.</p><p>Bias has as many definitions as the context in which you use it, I picked these two:</p><blockquote>The way our past experiences distort our perception of and reaction to information, especially in the context of mistreating other humans.</blockquote><blockquote>Algorithm bias occurs when a computer system reflects the implicit values of the humans who created it.</blockquote><p>I started looking at my diagram through the lens of potential bias and it now looks like this</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hmyyi56xFh62ydXdiBi86w.png" /></figure><p>If the top uses for conversational AI are entertainment and automation, what happens when the enterprise adoption turns a loan assessment into a conversational agent? If we continue to drill down, what happens if it uses <a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">a risk assessment API or model with a bias</a> to conduct the evaluation?</p><p>Do the same questions apply to a conversational agent for news feed? What happens when my smart speaker or phone curates Twitter trends but filter the content as it’s <a href="https://peopleofcolorintech.com/articles/how-automated-tools-discriminate-against-black-language/">considered hate or inappropriate </a>language?</p><h3>Why aren’t the big tech companies fixing it?</h3><p>As I kept discovering more and more studies and articles denouncing bias and calling out for changes and improvements, my first thought was “what are Amazon, Google, IBM, or Microsoft doing?”. And the answer is they are trying (<a href="https://developers.google.com/machine-learning/fairness-overview">Google</a>, <a href="http://aif360.mybluemix.net/">IBM</a>, <a href="https://www.microsoft.com/en-us/research/theme/fate/">Microsoft</a>), but they have three challenges to overcome:</p><ul><li>Despite their best efforts, with <a href="https://datawrapper.dwcdn.net/QojNb/1/">a limited diverse workforce</a>, these organizations will have difficulty correcting an issue that may have been caused indirectly by a lack of representation in the first place.</li><li><a href="https://www.technologyreview.com/2019/02/04/137602/this-is-how-ai-bias-really-happensand-why-its-so-hard-to-fix/">The challenges to solve </a>are complex: In our example earlier of the issues with the understanding of African American voices by speech to text engines, one tentative solution would be to retrain the language models with thousands of African American voice samples. It is <a href="https://voice.mozilla.org/en">not impossible</a>, but it takes time, and unfortunately, sometimes the data is not available.</li><li>No one solution can fix all the issues: what works for mitigating the bias in <a href="https://www.ted.com/talks/joy_buolamwini_how_i_m_fighting_bias_in_algorithms?language=en">facial recognition</a> may not apply to the struggles in conversational AI.</li></ul><h3>Top strategies for practitioners to mitigate bias</h3><p>I started thinking about how the product builders can help. The first thing I noticed was that even though we are using pre-built building blocks, these systems still allow us to do additional training specific to the product that we are working on.</p><p>Google or Alexa, for example, allows you to define language models on their platforms. One of the things you provide is called <a href="https://developer.amazon.com/en-US/docs/alexa/custom-skills/best-practices-for-sample-utterances-and-custom-slot-type-values.html">sample utterances</a> or <a href="https://cloud.google.com/dialogflow/docs/intents-training-phrases">training phrases</a>. This helps the system better understand when people are speaking to it.</p><p>Coming back to our earlier examples of poor speech to text, if we take an example from <a href="https://fairspeech.stanford.edu/">https://fairspeech.stanford.edu/</a>, I found two solutions to mitigate the issues of speech to text and natural language understanding.</p><h3>Mitigate the speech-to-text bias</h3><p>When doing high fidelity prototype user testing, using, for example, the Amazon Echo Show, with the team, we would review the Voice history (i.e. the transcript and the recording of what your Alexa enabled device heard).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FL2eE98F9oaNNclRGJGSRw.png" /><figcaption>Voice history of my attempt to set a timer for beans</figcaption></figure><p>This would enable us, during the testing, to find out whether Amazon (or any other voice-enabled devices) speech to text was having issues with some accents.</p><p>In the example above I want to say “<em>set beans timer for eight hours</em>”. For an application, I could add all the wrong transcripts (for example “<em>set beats timer for eight hours</em>”) to the sample utterances. Even though that’s not solving the root of the problem, it helps mitigate the issue for the users who are victims of the biased text to speech, because Amazon will not just end the conversation or provide the wrong response.</p><h3>Mitigate the natural language understanding bias</h3><p>In the Amazon console, in the <em>build</em> section, there is a <a href="https://developer.amazon.com/en-US/docs/alexa/custom-skills/review-intent-history-devconsole.html">very useful tool</a> that is a little hidden but that gives you access to a sample of Natural Language Understanding results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mXSh69dV8vGWnv-60xmixA.png" /></figure><p>Under the unresolved utterances are shown the utterances that users said but that could not be mapped to an intent. We usually mapped them to the closest intent. As with speech to text before, this will not entirely solve the issue but it provides a good mitigation solution.</p><p>This is only the beginning of the journey, j<a href="https://voicetechglobal.com/event/aaddressing-implicit-bias-against-the-black-community-in-conversational-products/">oin us</a> on our July 09th event to find out more strategies for product builders to mitigate the bias in conversational AI products.</p><h3>Acknowledgments</h3><p>A big <strong>THANK YOU</strong> and shoutout to <a href="https://www.linkedin.com/in/polina-cherkashyna-0a632758/">Polina Cherkashyna</a>, <a href="https://www.linkedin.com/in/aimee-reynolds-a259a619b/">Aimee Reynolds</a>, and <a href="https://www.linkedin.com/in/corin-lindsay-9388341a/">Corin Lindsay</a> for editing, reviewing, and ideating with me on this.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2012d822da95" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/how-product-builders-can-mitigate-bias-in-conversational-ai-2012d822da95">How Product Builders Can Mitigate Bias In Conversational AI</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lessons From Early Adopters: Best Voice First Products In Healthcare, Finance and Automotive]]></title>
            <link>https://medium.com/voice-tech-global/lessons-from-early-adopters-best-voice-first-products-in-healthcare-finance-and-automotive-201eba7f5813?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/201eba7f5813</guid>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Sat, 23 Feb 2019 17:35:32 GMT</pubDate>
            <atom:updated>2019-02-25T18:35:17.230Z</atom:updated>
            <content:encoded><![CDATA[<h3>Lessons From Early Adopters of Voice</h3><h4>Best Voice-First Products In Healthcare, Finance and Automotive</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x4pxphXMmHAp9iny5iFrgA.png" /></figure><p>You’ve read the headlines:<strong> voice is going to be big </strong>(and if you haven’t feel free to check our <a href="https://medium.com/voice-tech-global/smart-speakers-and-market-analysis-f0c418be08e5">smart speaker analysis</a>). Consumers will soon be surrounded by the technology and it could radically transform how they interact with the Internet and the spaces around them.</p><p>Google claims, for instance, a footprint of <strong>one billion devices</strong> powered by Google Assistant. Comscore -a digital audience measurement company- is forecasting that, by next year, <strong>50 per cent of searches</strong> will be conducted by voice.</p><blockquote>50 % of searches will be conducted by voice.</blockquote><p>If you’re a chief digital officer, brand manager or product designer, you must be wondering: What does it mean for my business? How will my industry be impacted? Who could I learn from?</p><p>To help you answer these questions here is an overview of interesting voice applications from the three <strong>major industries, which are successfully adopting voice</strong>: healthcare, finance and automotive.</p><h4><strong>Fastest Growing Applications Of Voice</strong></h4><p>According to an <a href="https://medium.com/point-nine-news/voice-tech-landscape-150-startups-mapped-and-analysed-82c5adaf710">analysis</a> by Point Nine Capital, a Berlin-based venture capital firm, healthcare is far and away the leading sector to experiment with voice solutions followed equally by finance, ecommerce and manufacturing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kcNOURl9p4C6XVGrcvNkTg.png" /><figcaption>Voice Tech Application Breakdown By Industry (Point Nine Capital)</figcaption></figure><p>Point Nine notes the rise of telemedicine and U.S. public policy changes that encourage follow-up care to minimize readmission are likely contributing factors to the adoption of voice technology. In finance, the rapid end of in-person banking has pushed financial institutions to find new ways to maintain strong client relationships through engagement on multiple platforms in an effort to avoid growing customer churn rates.</p><p>Here are the most interesting voice applications that might offer lessons to business leaders across sectors.</p><h4><strong>Healthcare</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tVYlWfXwuOw7PPLGXFmtWQ.png" /></figure><p>At the Consumer Electronics Show this year, SameDay Security announced their virtual caregiver application <a href="http://ces.vporoom.com/2018-12-27-Electronic-Caregiver-Announces-Addison-Care-TM-The-Worlds-First-Virtual-Caregiver-TM-the-Most-Ambitious-Tech-at-CES">Addison Care</a>.The company already offers a suite of products for elderly care. But this ambition initiative integrating voice, 3D animated video, machine learning and more connected through multiple devices and technologies could very well become the gold standard of in-home care for our rapidly aging population.</p><p>A more modest, yet accessible, solution is available right now to owners of a Google Home or Amazon Alexa. <a href="https://cuidahealth.com/lisa/"><strong>Learning Interface for Senior Adults</strong></a>(LiSA) by Cuida Health takes a number of features like routines, reminders, local event calendar, and hands-free group messaging, and packages them in an application tailored to senior-aged users. This assistant application empowers seniors to continue living independently and helps keep them connected with others in their community encouraging its users to be social. Said customer Jim Kitzmiller of the Alexa skill:</p><blockquote>Lisa is a wonderful companion for seniors. She makes sure I take my meds. She keeps me informed. And she keeps getting smarter all the time.</blockquote><p><a href="https://askmarvee.com/">Marvee</a>, a three-year old voice-first agency, has launched a number of voice applications geared at senior care. Inspired by its founder Heidi Culbertson’s personal experience with hands-free solutions to improve the quality of life of her aging mother, their studio has launched a small suite of applications in that vein.</p><p><a href="https://www.amazon.com/Chair-Workout-Sit-Stay-Fit/dp/B07K4F157J"><strong>Chair Workout — Sit and Stay Fit</strong></a> encourages seniors to be active. It offers three workouts for people with limited mobility which are meant to be done while seated. One reviewer said of the Chair Workout:</p><blockquote>My Mom is in a nursing home and for Christmas she got Alexa. Setting this skill up in her morning routine will get her ahead of the game. She’s been bedridden a year because of her lower legs. The skill allows her to focus on just the areas she had good control over. The commands are simple, yet valuable. The skill includes a little humor, some motivation and I’m very happy with its performance. Thank you skill writer for this thoughtful, well executed skill.</blockquote><p><a href="https://www.amazon.com/Marvee-LLC-Senior-Moments/dp/B07HWNF4YC/"><strong>Senior Moments</strong></a> is an entertainment skill offering a mix of jokes, facts and wisecracks about aging. One reviewing said of the skill</p><blockquote>Great quotes. All ages can relate!</blockquote><p>In the realm of patient care, <a href="https://www.amazon.com/Boston-Childrens-Hospital-KidsMD/dp/B01E3F0AUC"><strong>KidsMD</strong></a> from Boston Children’s Hospital is an application geared at parents looking for general answers to common medical questions from a trusted source. While not intended as a substitute for professional medical advice, its utility can potentially ease the burden on the ER or drop-in clinics through reduced visits.</p><h4><strong>Financial</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*fcHMbJsUOXno-vWCGhGpoQ.png" /></figure><p>Finance is an extremely competitive industry. Banks are facing disruption from both ends: on the one side, upstart digital-native banking alternatives crowding the market; on the other, rapidly changing consumer behaviour as people embrace digital and mobile banking for everything from day-to-day personal needs to trading stocks. Making innovation a bit more challenging is the public policy regulating what financial institutions can and cannot do with people’s personal data as it relates to their accounts.</p><p>Nevertheless, companies like PayPal, Amex, Capital One, and Fidelity have found ways to offer customers a way to interact through voice with their account information or the companies’ services. Of note, the <a href="https://www.amazon.com/Ally-Financial-Inc/dp/B07763RXZJ"><strong>Ally</strong></a> skill by Ally Financial Inc. The app allows customers the ability to review their account balance and recent account activity as well as transfer funds. Many customers are surprised by its usefulness. Customer Daniel writes in his review:</p><blockquote>Didn’t realize that I would use this frequently. The balance function is very helpful, need not use my mobile anymore!</blockquote><p>And reviewer Ross writes:</p><blockquote>The entire setup process was a breeze. I never dreamt of using this so frequently. Great experience using the skill. Love it!</blockquote><p><strong>Automotive</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*M88DtbC-EzJvMJNzdQBzfA.png" /></figure><p>It’s going to be a huge market and <a href="https://www.tractica.com/user-interface-technologies/in-car-assistants-bring-voice-ai-to-the-masses/">automakers have already rolled out the technology with some 2018/2019 models</a>. Toyota introduced Alexa controls in select Toyota and Lexus models this past year. BMW is also working with Amazon and should be on the market shortly; they’ve also struck a R&amp;D deal with IBM. Honda, Hyundai and GM are working on Android Auto solutions to offer Google Assistant. Additionally, players like Nuance are developing AI technology married with voice activated services such as Dragon Drive to power connected cars.</p><p>If 2018 was the year of the first movers, the next 12 months will provide us with the opportunity to observe the results of early adopters’ tests and learn how users engage with their applications. These insights should help brands like yourself understand ways to connect with your customers on voice-enabled platforms, and will help product designers like us further refine the types of high value voice solutions we build for clients.</p><h4><strong>Did You Like This Article?</strong></h4><p>We hope that you found this post useful and we’d love to hear your feedback! Please <strong>support us with some claps</strong> and join the conversation on <a href="https://twitter.com/voicetechglobal"><strong>Twitter</strong></a>, <a href="https://bit.ly/2DE0BD1"><strong>Slack</strong></a> or<strong> </strong><a href="https://www.linkedin.com/company/voice-tech-to/"><strong>LinkedIn</strong></a>. And if you’re interested in getting hands-on with voice look out for out <a href="https://www.meetup.com/voice-tech-to/"><strong>Meetup events and webinars</strong>!</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=201eba7f5813" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/lessons-from-early-adopters-best-voice-first-products-in-healthcare-finance-and-automotive-201eba7f5813">Lessons From Early Adopters: Best Voice First Products In Healthcare, Finance and Automotive</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Looking back at the Voice Summit today]]></title>
            <link>https://medium.com/voice-tech-global/looking-back-at-the-voice-summit-today-f0af912ddfde?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0af912ddfde</guid>
            <category><![CDATA[voice-application]]></category>
            <category><![CDATA[voice-summit]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[technology-news]]></category>
            <category><![CDATA[voice-technology]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Sun, 16 Dec 2018 19:30:38 GMT</pubDate>
            <atom:updated>2019-02-23T18:17:02.491Z</atom:updated>
            <content:encoded><![CDATA[<h3>Looking Back at the Voice Summit</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/384/1*YvcE1pDieF2pg38OboJPAg.gif" /></figure><p>Back in July, Polina, Tim and I attended the Amazon-sponsored Voice Summit in Newark, NJ.</p><p>The event was attended by more than 200 speakers and 3,000 conference-goers. If you want to experience much of the content from the event, there are <a href="https://www.youtube.com/playlist?list=PLn51IO3rbkV1E1a6WjgvFtW3VaOCRxzov">165 videos</a> available. Be prepared, however, as many of the videos run 30 minutes or longer.</p><p>With the benefit of six months since attending the summit, we’d like to offer our reflections and share a few key themes we extracted during two very full days densely filled with presentations and workshops we think paint a picture of the Voice technology space as we know it today.</p><p>[To read roundups from the conference, <a href="https://www.google.ca/search?q=takeaway+from+the+voice+summit+in+newark&amp;oq=takeaway+from+the+voice+summit+in+newark">follow this link</a> to a Google search results list.]</p><p>Over the course of a three-part series, we’ll look at the global state of Voice technology — the current major players in the space as well as some of the remaining challenges that prevent the technology from becoming an established worldwide channel for services such as interactive voice response. Then, we do a deep-dive into specific industries such as healthcare and automotive and the impact offered by voice user interfaces, exploring current use cases as well as highlight some key points for future opportunities. Finally, we will drill down to the practitioners level by discussing the design as the differentiator for the technology’s future.</p><p><a href="https://medium.com/@gtonye/smart-speakers-and-market-analysis-f0c418be08e5">Part I: Smart speakers &amp; market analysis</a></p><p><a href="https://medium.com/voice-tech-global/lessons-from-early-adopters-best-voice-first-products-in-healthcare-finance-and-automotive-201eba7f5813">Part II: Lessons From Early Adopters of Voice</a></p><p>Part III: Building Voice application (Link coming soon)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oKALypHHqeggIm-rG0h5Qw.jpeg" /><figcaption>(from left to right) Guy, <a href="https://medium.com/@pollycherkashyna">Polina</a> and <a href="https://medium.com/@tim_bettridge">Tim</a></figcaption></figure><p>We also wanted to give a shout out to the city of Newark. While you did not end up winning the bid for Amazon HQ2, we received an amazing welcome from you during the summit. For those of you who travel through Newark on your way to New York City and other destinations, there is a lot more to see than just the airport.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f0af912ddfde" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/looking-back-at-the-voice-summit-today-f0af912ddfde">Looking back at the Voice Summit today</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Smart Speakers Market Analysis]]></title>
            <link>https://medium.com/voice-tech-global/smart-speakers-and-market-analysis-f0c418be08e5?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0c418be08e5</guid>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[voice-tech]]></category>
            <category><![CDATA[market]]></category>
            <category><![CDATA[strategy]]></category>
            <category><![CDATA[technology-news]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Sun, 16 Dec 2018 19:05:35 GMT</pubDate>
            <atom:updated>2019-01-02T20:53:03.744Z</atom:updated>
            <content:encoded><![CDATA[<h3>Smart Speaker Market Overview</h3><p>With the holiday season around the corner, it’s expected that smart speakers will be another hot ticket item on people’s wish lists.</p><p>In this, our first article in <a href="https://medium.com/voice-tech-global/looking-back-at-the-voice-summit-today-f0af912ddfde">a series looking at key themes from Voice Summit 2018</a>, we will explore consumer adoption of smart speakers and which companies are leading the market. While voice technology is not hardware specific, most of the industry currently tends to focus on those devices as they are becoming increasingly popular in households. And holiday sales will continue to be a measure of success.</p><h4>User behaviour</h4><p><a href="https://www.youtube.com/watch?v=MCYOGHLQbS4">In one of the first talks at Voice Summit, Brett Kinsella of VoiceBot.ai</a>, a one-stop shop for “important news, commentary, research and analysis of voice technology,” shared a number of interesting industry insights.</p><p><strong>Smart speakers adoption is growing at a quicker pace than previous technologies</strong>. Kinsella’s Voicebot report indicated that, by March 2018, approximately 19% of U.S. adults owned a smart speaker. By comparison, it took 13 years for television to match that same penetration in the U.S. The growth is closer to how the social networks are expanding.</p><p><strong>Smart speakers are not idle in the household</strong>. An interesting data point from their November report is that smart speakers owners are interacting regularly with their device: 45 per cent daily and 79 per cent monthly. You can check out more smart speaker data <a href="https://voicebot.ai/download-smart-speaker-consumer-adoption-report-2018/">in their 2018 “Adoption Report.”</a></p><p><em>What might explain the popularity of these devices</em>?</p><p>Well, onboarding for voice-enabled devices appears to be much easier compared to other mediums like a computer or smartphone. Despite amazing efforts from UX, UI and graphic designers, sometimes it is not the easiest task to onboard a user to a new application on platforms such as a screen with a remote, a smartphone or a tablet. On the other hand, voice appears to be much more natural. People have been doing it for tens (possibly hundreds) of thousands of years all over the world. This differs from understanding a touchscreen or using a mouse, which could be a barrier preventing some users from embracing the technology.</p><p>Another aspect to look at when analyzing the success of smart speakers is how its owners quickly become promoters. Kinsella illustrated this point by asking the audience: <em>how many people have used a smart speaker with another person in the room</em>? Watching someone interact with the technology is more powerful even than word of mouth when it comes to convincing someone to pick up a new technology.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3nXRVIJkjUWC_eFqp9169Q.png" /><figcaption>source: <a href="https://voicebot.ai/2018/07/25/smart-speakers-are-the-gateway-to-increased-voice-assistant-use-on-mobile/">Voicebot.ai</a></figcaption></figure><p>Considering its adoption rate, we seem to have moved past the question of “Is voice a thing?” into the domain of “Where do I get started?” Businesses and product builders should not wonder whether or not to play in the voice space but focus, rather, on how to add it in their value proposition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*g1iS4Stopsp726GwX_IIBA.jpeg" /></figure><h4><strong>Who is leading the voice assistant war?</strong></h4><p>Amazon, the pioneer of smart speakers which are built on solid AWS infrastructure, is ahead of the pack at 31.9% of global smart speaker sales. Their dominance is mainly due to the ability to ride each wave of new internet trends. They were early to set the standard in e-commerce transactions and early to design search in the product catalog or recommendation. These trends, coupled with a large customer base on their two-click two-day marketplace, supports their top rank.</p><p>At 29.8%, Google is second but catching up to Amazon, positioning themselves to be able to leverage the vast user base for their many existing products including Search, G Suite and Android.</p><p>Apple is a distant third with Siri. While it offered the first voice assistant to the market through iPhones, its capabilities underwhelmed many users. Their HomePod has failed to grab many of the Apple faithful. But a leap in Siri’s technology could turn the market upside down. We’d consider it a sleeper brand in this industry.</p><p>Finally, Microsoft has been sharpening its voice assistant technology, dubbed Cortana, but has not made a strong push yet. The opportunity remains considerable given their strong presence in the enterprise market to leverage voice to help power successful services such as Microsoft Office.</p><p>Other potential contenders include Samsung. Although not an active player, it is one of the leaders in hardware and with a sizeable ownership of the mobile market, its virtual assistant, Bixby, could become a rival. There is also IBM. Although the strategy with Watson is not clear yet, its enterprise business could be a launch pad into the voice market. Finally we have Facebook who has been leading the charge with chatbot technology and its assistant M. Despite enabling their <a href="https://portal.facebook.com/">Portal</a> suite of video chat devices with Alexa, their two billion global user base make them a threat to any leader in the voice space.</p><p>The state of the global market is also split along the Easter and Western hemispheres. Aside from the key North American players we’ve looked at, a lucrative market like China, for instance, is dominated by home-grown companies including Xiaomi with <a href="https://www.youtube.com/watch?v=6iiiYI08dX4">Xiao Ai</a>, Baidu Dueros and Alibaba (<a href="https://www.technologyreview.com/s/612511/alibaba-already-has-a-voice-assistant-way-better-than-googles/">recently praised for a stunning demo of its own Duplex</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O9sR7kivutl5N1EQe22g2Q.png" /><figcaption>source: <a href="https://voicebot.ai/2018/11/16/china-jumps-to-29-of-smart-speaker-sales-in-q3-2018-u-k-hits-5-and-the-u-s-falls-to-42/">voicebot.ai</a></figcaption></figure><p>Despite the current leaderboard ranking, the race is far from over.</p><p><strong>Users are not yet locked to a specific platform</strong>. However, Alexa and Google Assistant have the greatest awareness among consumers.</p><p><strong>Virtual assistants need to become very performant</strong>. Amazon and Google are at an advantage here as their assistants have been trained over the past few years with extensive polishing of Natural Language Processing and Natural Language Understanding.</p><p>Despite growing consumer adoption, a few more hurdles need to be cleared. For instance, there are still untapped non-English markets around the world, but internationalizing the technology is a bit more challenging than simply translation as cultural context — the nuances and specificities in language and conversation — is an important aspect to voice. Security and privacy are also of concerns to consumers to a greater or lesser degree depending on the market.</p><p>If you were at the summit, let us know what we may have missed in terms of introducing smart speakers and the status quo of Voice technology.</p><p><a href="https://medium.com/@pollycherkashyna">Polina</a>, <a href="https://medium.com/@tim_bettridge">Tim</a> and Guy</p><p>p.s. A special shout out to <a href="https://www.linkedin.com/in/jessie-j-sun-design/">Jess</a> for taking the time to edit this article.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f0c418be08e5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/voice-tech-global/smart-speakers-and-market-analysis-f0c418be08e5">Smart Speakers Market Analysis</a> was originally published in <a href="https://medium.com/voice-tech-global">Voice Tech Global</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[3 Takeaways from Alexa Dev Day]]></title>
            <link>https://medium.com/connected/3-takeaways-from-alexa-dev-day-b1e4c8455147?source=rss-11382a5b2c64------2</link>
            <guid isPermaLink="false">https://medium.com/p/b1e4c8455147</guid>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[alexa-skills]]></category>
            <category><![CDATA[alexa]]></category>
            <category><![CDATA[conversational-ui]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Guy TONYE]]></dc:creator>
            <pubDate>Mon, 05 Mar 2018 19:23:21 GMT</pubDate>
            <atom:updated>2018-03-05T19:56:51.646Z</atom:updated>
            <content:encoded><![CDATA[<h4>Key tools and practices for the creation of effective Alexa skills</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5wd0zDlJisc7sZ69u2V-2A.png" /></figure><p>As a software developer based out of Toronto (and one of the organizers of <a href="https://www.meetup.com/voice-tech-to">Voice Tech TO</a>), I recently took a trip down to NYC to attend one of Amazon’s infamous <a href="https://developer.amazon.com/alexa/devday">Alexa Dev Days</a>. For those unfamiliar with the event, Alexa Dev Days are recurring, multi-city developer workshops that are free and open to the public. Full of helpful info, they are geared toward the design and development of third-party skills for Alexa-enabled devices.</p><p>It’s quite a trip from Toronto to New York just to take a one-day workshop, but looking back on all the things I learned and discovered, it was 100% worth it.</p><p>In a nutshell, the event covered three major topics:</p><ol><li><em>How to Design a Conversation</em></li><li><em>An Overview of Amazon’s Developer Tools</em></li><li><em>New Features for Alexa</em></li></ol><p>Let’s go through these one by one.</p><h3>1. How to Design a Conversation</h3><p>Since designing for the ear is so different from designing for the eye, the Alexa team conducted a number of small exercises to get people into the right headspace.</p><p><strong>First Exercise<br></strong>After grouping us into teams, the Alexa instructors challenged us to design a simple recommendation skill. We had to figure out: what does our skill recommend, exactly? And how, conversationally, do users arrive at those recommendation(s)?</p><p>My team chose to build a bike recommendation engine.</p><p>As we learned, everything begins with writing the conversation in its ideal form, commonly referred to as the “happy path.” Using nothing but plain old paper, this is the dialogue we came up with:</p><blockquote>User (U): Open bike assistant.<br>Alexa (A): Hi, I can help you find a bike. Are you looking for a city bike or a mountain bike?<br>(U): A city bike.<br>(A): Okay. How often do you ride your bike on average?<br>(U): Twice a week.<br>(A): Got it. What is your budget?<br>(U): Around $200.<br>(A): Okay. Searching… I would recommend a carbon road bicycle from Amazon at $220. Would you like me to send the information to your phone?<br>(U): Yes.<br>(A): Sending now.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ziC61IIhZDItbwhBOJRXNg.jpeg" /></figure><p><strong>Second Exercise<br></strong>On the basis of our happy path, our next exercise was to formally document the user’s overall intention, as well as the kinds of information necessary for the skill to <em>satisfy</em> that intention, aka the different input types, or <strong>slots</strong>. Finally, we needed to identify the actual inputs that could legitimately be entered under each of these slots, aka the <strong>values</strong>. As you’ll see in the next section, we needed to do this because, when you build the actual skill, you can’t just type up and submit a transcript of your desired conversation. You need to build the program in terms of intentions, slots, and legitimate value entries.</p><blockquote><strong>User Intention: </strong>An appropriate bike recommendation<br><strong>Slots: </strong>Bike type, frequency, budget<br><strong>Values: </strong>City Bike or Mountain Bike (for Bike Type); Once a week, twice a week, daily, or often (for the Ride frequency); Any number (for Budget)</blockquote><blockquote><strong>User Intention</strong>: Receiving the recommendation on your phone<br><strong>Slot: </strong>Preference<br><strong>Values</strong>: Yes, No</blockquote><p>At this point in the exercise, the Alexa team emphasized the need to perform and test our skill with real users in order to learn more about their intentions and add/remove slots and values accordingly.</p><h3>2. Amazon Developer Tools: An Overview</h3><p>After designing your happy path and identifying the relevant slots and values, the workshop’s focus shifted toward actually developing the skill. The Alexa team walked us through a number of tools to help with that, loosely divided into “graphical” tools and more traditional “console” tools.</p><h4>Graphical Tools</h4><p>“Graphical” skill building, remarkably, does not require any coding, occurring entirely in your web browser using a combination of the <a href="https://developer.amazon.com/alexa/console/ask">Alexa Skills Kit Developer Console</a>, <a href="https://alexa.design/skillcode">Skill Code Generator</a>, and <a href="https://console.aws.amazon.com/">AWS Lambda Console</a> (don’t be misled by the word “console” here — these are GUIs). Let’s go through these one by one.</p><p><a href="https://developer.amazon.com/alexa/console/ask"><strong>Alexa Skills Kit Developer Console</strong></a><strong><br></strong>Many of you may already know this tool, though it got a pretty solid facelift. Basically, it allows developers to define the intention, slots, and values using its GUI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jS08jhbiVYHmAuaMoh-fIA.png" /><figcaption>Defining the slots: bikeType, bikeRidingFrequencyType, and budget.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UdTnAfziwFh6hycwuun6Dw.png" /><figcaption>Defining the values for bikeRidingFrequencyType, with synonyms.</figcaption></figure><p>Notice in the above image that there’s a column for synonyms: for each slot, you can define values and add synonyms that would resolve to the same value.</p><p>You can also quickly define what information is mandatory for each utterance and automate the question that will be prompted to the user if that info is missing (below).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6UO0C6oWOcwYSAAGb-RhiQ.png" /><figcaption>BikeType is defined as a mandatory input for the recommendation, so that if it is not provided already, Alexa will automatically prompt the user.</figcaption></figure><p><a href="https://alexa.design/skillcode">Skill Code Generator</a><br>Skill Code was one of the most impressive tools presented during the event. After building everything in the Alexa Skills Kit Developer Console and receiving a model of your skill (in the form of a json file), Skill Code lets you copy it into a little console which automatically generates the code for that skill in JavaScript. One click and your index.js is ready for download and use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tEFoDQYo_OWFfZQ7." /><figcaption>The code on the right is generated from the skill’s model json and is ready for use in the Adapter Lambda.</figcaption></figure><p><a href="https://console.aws.amazon.com/">AWS Lambda Console</a><br>AWS Lambda Console lets you define the <a href="https://developer.amazon.com/docs/custom-skills/host-a-custom-skill-as-an-aws-lambda-function.html">lambda</a> that will fulfill the skill’s request — in our case, finding an appropriate bicycle recommendation from the immense sea of bikes out there. Then the previously generated index.js can be uploaded to your lambda <em>et</em> <em>voilà,</em> the skill is ready for you to try out in Alexa Skills Kit Developer Console or through a configured device.</p><h4>Console</h4><p>If GUIs aren’t your favourite, or if versioning and continuous deployment are necessary for your project (they often are), an old-fashioned terminal is the choice for you. For the developers in the room, Amazon presented their command line “ask-cli.” It allows builders to manage skill configuration directly from the console.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/665/0*qrmP-TEHWPVv6SeG." /></figure><p>Another benefit of ask-cli is that it gives developers tools for lambda management that were previously only available via non-Amazon software such as <a href="http://apex.run/">Apex</a> and <a href="https://serverless.com/">Serverless</a>. (More information <a href="https://developer.amazon.com/docs/smapi/quick-start-alexa-skills-kit-command-line-interface.html">here</a>.)</p><p>Whether you’re a designer or a developer, a product manager or simply a voice enthusiast, Amazon provides the tools for you to make a skill.</p><h3>3. Alexa’s New Features</h3><p>Another interesting moment of the day was the presentation of some upcoming features Amazon had in store for Alexa, relevant for developers no less than users.</p><h4>Gadget API</h4><p>Amazon is offering a new type of voice experience through an extended product line they’re calling Gadget. The first product available is called <a href="https://www.amazon.com/Echo-Buttons-Alexa-Gadget-Pack/dp/B072C4KCQH">Echo Button</a> and it was released in the US in December 2017. It is meant to enhance the gaming experience on Alexa, with a trivia game for example, allowing users to race each other to answer questions. An <a href="https://developer.amazon.com/alexa/alexa-gadgets/gadgets-skill-api">API</a> allowing third-party skills to interact with these gadgets will be made available soon.</p><h4>Progressive Response</h4><p>When a user makes their request, depending on the network or the query processing, the user could end up waiting for a significant amount of time. As of November 2017, Alexa now supports <a href="https://developer.amazon.com/docs/custom-skills/send-the-user-a-progressive-response.html">progressive response</a>, which allows skills to send intermediate responses to the user while processing the real request.</p><h4>Alexa for Business</h4><p>As of November 2017, AWS is offering a series of features around <a href="https://aws.amazon.com/alexaforbusiness/">Alexa for Business</a>. It provides a suite of management tools (like fleet management for company devices) and easily integrates with enterprise software (like conferencing and help desk). Developers also have the opportunity to build skills for corporations in a private skill store using a wider range of <a href="https://docs.aws.amazon.com/a4b/latest/APIReference/Welcome.html">APIs.</a></p><h4>Notifications</h4><p>Currently, Domino’s pizza is the only service that can notify users on Alexa. However, a notification option for all skills are in the works. It’s a <a href="https://developer.amazon.com/blogs/alexa/post/833b9af4-26e6-47d2-a13d-bdbd9a257512/expanded-developer-preview-of-notifications-for-alexa-skills">sensitive topic</a>, but a <a href="https://alexa.au1.qualtrics.com/jfe/form/SV_72lplGGegxNZ9Ln">developer preview</a> will be available soon.</p><h4>Voice ID</h4><p>Voice recognition is already available for users on Alexa, but soon developers will be able to use it <a href="https://developer.amazon.com/blogs/alexa/post/09acb84e-6d56-463b-959c-9e15c3ed3893/announcing-more-personalized-skill-responses-when-alexa-recognizes-a-customer-s-voice-developer-preview">to provide customized experiences</a>. Stay tuned <a href="http://dev.amazonappservices.com/alexa-voice-profile.html">here</a>.</p><h4>SearchQuery slot</h4><p>To leverage less predictable user utterances, Alexa added a SearchQuery slot-type to its Skill Kit that can be applied to specific slots to allow for open-ended inputs/values to simplify the design of a skill. See an example <a href="https://developer.amazon.com/docs/custom-skills/slot-type-reference.html#amazonsearchquery">here</a>.</p><h4>IoT</h4><p>Though not new, the team showed how integrating Alexa Voice Service (AVS) into a device allows you to interact with it via Alexa. To illustrate their point, they showed how Alexa could be used to control your web browser to find images quickly online (source code <a href="https://github.com/alexa/alexa-cookbook/tree/master/aws/Amazon-IOT/webapp-thing">here</a>).</p><h3>Conclusion</h3><p>After a jam-packed day at the W Hotel in midtown Manhattan, the strongest impression I was left with is that Amazon is committed to sharpening their skill-building tools for designers and developers alike, no doubt in order to boost production and provide their users with the best experiences possible. Smart.</p><p>If you don’t have a chance to attend one of these Dev Days, feel free to reach out or come to the next meeting of <a href="https://www.meetup.com/voice-tech-to/">Voice Tech TO</a>. A Toronto-based event that explores Google Home, Amazon Alexa, and the wide world of voice, it’s our own personal Dev Day ;-)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*uvLyWQjwYHgt4BKxgOcssw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1e4c8455147" width="1" height="1" alt=""><hr><p><a href="https://medium.com/connected/3-takeaways-from-alexa-dev-day-b1e4c8455147">3 Takeaways from Alexa Dev Day</a> was originally published in <a href="https://medium.com/connected">Connected</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>