{"id":677448,"date":"2019-03-05T09:00:14","date_gmt":"2019-03-05T17:00:14","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&#038;p=677448"},"modified":"2020-07-22T11:08:12","modified_gmt":"2020-07-22T18:08:12","slug":"guidelines-for-human-ai-interaction-eighteen-best-practices-for-human-centered-ai-design","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/guidelines-for-human-ai-interaction-eighteen-best-practices-for-human-centered-ai-design\/","title":{"rendered":"Guidelines for Human-AI Interaction: Eighteen best practices for human-centered AI design"},"content":{"rendered":"<p>By <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/www.linkedin.com\/in\/mihaelavorvoreanu\/\">Mihaela Vorvoreanu<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/www.linkedin.com\/in\/saleema-amershi\/\">Saleema Amershi<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, and <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/www.linkedin.com\/in\/pennycollisson\/\">Penny Collisson<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n<div id=\"attachment_677451\" style=\"width: 1060px\" class=\"wp-caption alignnone\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-677451\" class=\"wp-image-677451 size-full\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2020\/07\/Guildelines-for-AI_header.jpeg\" alt=\"Graphic artwork. In the foreground, two hands hoover over a computer keyboard, preparing to type. In the background, various sized triangles and shapes float across to create a design.\" width=\"1050\" height=\"559\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2020\/07\/Guildelines-for-AI_header.jpeg 1050w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2020\/07\/Guildelines-for-AI_header-300x160.jpeg 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2020\/07\/Guildelines-for-AI_header-1024x545.jpeg 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2020\/07\/Guildelines-for-AI_header-768x409.jpeg 768w\" sizes=\"auto, (max-width: 1050px) 100vw, 1050px\" \/><p id=\"caption-attachment-677451\" class=\"wp-caption-text\">Note: This article was originally published on March 5, 2019, on Medium<\/p><\/div>\n<p>Today we\u2019re excited to share a set of\u00a0<a class=\"bx dm lj lk ll lm\" href=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/blog\/guidelines-for-human-ai-interaction-design\/\" target=\"_blank\" rel=\"noopener nofollow noreferrer\">Guidelines for Human-AI Interaction<\/a>. These 18 guidelines can help you design AI systems and features that are more human-centered. Based on more than two decades of thinking and research, they have been validated through a rigorous study published in\u00a0<a class=\"bx dm lj lk ll lm\" href=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/publication\/guidelines-for-human-ai-interaction\/\" target=\"_blank\" rel=\"noopener nofollow noreferrer\">CHI 2019<\/a>.<\/p>\n<h3 id=\"06ae\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong>Why do we need guidelines for human-AI interaction?<\/strong><\/h3>\n<p id=\"256c\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">While classic interaction guidelines hold with AI systems, attributes of AI services, including their accuracy, failure modalities, and understandability raise new challenges and opportunities. Consistency, for example, is a classic design guideline that advocates for predictable behaviors and minimizing unexpected changes. AI components, however, can be inconsistent because they may learn and adapt over time.<\/p>\n<p class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">We need updated guidance on designing interactions with AI services that provide meaningful experiences, keeping the user in control, and respecting users\u2019 values, goals, and attention.<\/p>\n<h3 id=\"86aa\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong class=\"ay\">Why these guidelines?<\/strong><\/h3>\n<p>AI-focused design guidance is blooming across UX conferences, the tech press, and within individual design teams. That\u2019s exciting, but it can be hard to know where to start. We wanted to help with that, so\u2026<\/p>\n<ul class=\"\">\n<li id=\"1d7f\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn mt mu mv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">We didn\u2019t just make these up!<\/strong>\u00a0<strong class=\"kc mw\">They come from more than 20 years of work.\u00a0<\/strong>We read numerous research papers, magazine articles, and blog posts. We synthesized a great deal of knowledge acquired across the design community into a set of guidelines that apply to a wide range of AI products, are specific, and are observable at the UI level.<\/li>\n<li id=\"1910\" class=\"ka kb bj kc b hv mx ke hy my kg kh mz id kj na ig kl nb ij kn mt mu mv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">We validated the guidelines through rigorous research.<\/strong>\u00a0We tested the guidelines through three rounds of validation with UX and HCI experts. Based on their feedback, we iterated the guidelines until experts confirmed that they were clear and specific.<\/li>\n<\/ul>\n<h3 id=\"03d2\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong class=\"ay\">Let\u2019s dive into the guidelines!<\/strong><\/h3>\n<p>The guidelines are grouped into four categories that indicate when during a user\u2019s interactions they apply: upon initial engagement with the system, during interaction, when the AI service guesses wrong, and over time.<\/p>\n<h3 id=\"4133\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong>Initially<\/strong><\/h3>\n<h4 id=\"594d\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">1. Make clear what the system can do.<\/h4>\n<h4 id=\"1667\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">2. Make clear how well the system can do what it can do.<\/h4>\n<p id=\"a032\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">The guidelines in the first group are about setting expectations: What are the AI\u2019s capabilities? What level of quality or error can a user expect? Over-promising can hurt perceptions of the AI service.<\/p>\n<p id=\"083a\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">PowerPoint\u2019s QuickStarter<\/strong>\u00a0illustrates Guideline 1,\u00a0<em class=\"ko\">Make clear what the system can do<\/em>. QuickStarter is a feature that helps you build an outline. Notice how QuickStarter provides explanatory text and suggested topics that help you understand the feature\u2019s capabilities.<\/p>\n<p data-selectable-paragraph=\"\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-677472 size-full\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-2.png\" alt=\"Photo of Quickstarter. Top, text reads Example in Practice with search bar below\" width=\"406\" height=\"576\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-2.png 406w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-2-211x300.png 211w\" sizes=\"auto, (max-width: 406px) 100vw, 406px\" \/><\/p>\n<h3 id=\"b010\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong>During interaction<\/strong><\/h3>\n<h4 id=\"798d\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">3. Time services based on context.<\/h4>\n<h4 id=\"7ff7\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">4. Show contextually relevant information.<\/h4>\n<h4 id=\"e299\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">5. Match relevant social norms.<\/h4>\n<h4 id=\"0e33\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">6. Mitigate social biases.<\/h4>\n<p id=\"af43\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">This subset of guidelines is about context. Whether it\u2019s the larger social and cultural context or the local context of a user\u2019s setting, current task, and attention, AI systems should take context into consideration.<\/p>\n<p id=\"9a12\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\">AI systems make inferences about people and their needs, and those depend on context. When AI systems take proactive action, it\u2019s important for them to behave in socially acceptable ways.\u00a0To apply Guidelines 5 and 6 effectively,\u00a0<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/www.wsj.com\/articles\/a-crucial-step-for-avoiding-ai-disasters-11550069865?redirect=amp\" target=\"_blank\" rel=\"noopener noreferrer\">ensure your team has enough diversity to cover each other\u2019s blind spots<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>.<\/p>\n<p id=\"402c\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">Acronyms in Word<\/strong>\u00a0highlights Guideline 4, <em>S<\/em><em class=\"ko\">how contextually relevant information.<\/em>\u00a0It displays the meaning of abbreviations employed in your own work environment relative to the current open document.<\/p>\n<p data-selectable-paragraph=\"\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-677475 size-full\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-3.png\" alt=\"Screenshot of the Acronyms function in Microsoft Word. Text at top reads Example in Practice\" width=\"406\" height=\"576\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-3.png 406w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-3-211x300.png 211w\" sizes=\"auto, (max-width: 406px) 100vw, 406px\" \/><\/p>\n<h3 id=\"00f4\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong>When wrong<\/strong><\/h3>\n<h4 id=\"71c9\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">7. Support efficient invocation.<\/h4>\n<h4 id=\"e79e\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">8. Support efficient dismissal.<\/h4>\n<h4 id=\"5744\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">9. Support efficient correction.<\/h4>\n<h4 id=\"396c\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">10. Scope services when in doubt.<\/h4>\n<h4 id=\"f757\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">11. Make clear why the system did what it did.<\/h4>\n<p id=\"7352\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">Most AI services have some rate of failure. The guidelines in this group recommend how an AI system should behave when it is wrong or uncertain, which will inevitably happen.<\/p>\n<p id=\"ce13\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\">The system might not trigger when expected, or might trigger at the wrong time, so it should be easy to invoke (Guideline 7) and dismiss (Guideline 8). When the system is wrong, it should be easy to correct it (Guideline 9), and when it is uncertain, Guideline 10 suggests building in techniques for helping the user complete the task on their own. For example, the AI system can gracefully fade out, or ask the user for clarification.<\/p>\n<p id=\"dc3c\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">Auto Alt Text<\/strong>\u00a0automatically generates alt text for photographs by using intelligent services in the cloud. It illustrates Guideline 9,\u00a0<em class=\"ko\">Support efficient correction<\/em>, because automatic descriptions can be easily modified by clicking the Alt Text button in the ribbon.<\/p>\n<p data-selectable-paragraph=\"\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-677487 size-full\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-5.png\" alt=\"Screengrab of Alt Text example\" width=\"406\" height=\"576\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-5.png 406w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-5-211x300.png 211w\" sizes=\"auto, (max-width: 406px) 100vw, 406px\" \/><\/p>\n<h3 id=\"6702\" class=\"ln lo bj bi fs lp lq lr ls lt lu lv lw lx ly lz ma mb mc md me cx\"><strong>Over time<\/strong><\/h3>\n<h4 id=\"92d0\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">12. Remember recent interactions.<\/h4>\n<h4 id=\"4d3a\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">13. Learn from user behavior.<\/h4>\n<h4 id=\"b305\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">14. Update and adapt cautiously.<\/h4>\n<h4 id=\"13b9\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">15. Encourage granular feedback.<\/h4>\n<h4 id=\"1846\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">16. Convey the consequences of user actions.<\/h4>\n<h4 id=\"8908\" class=\"nd lo bj bi fs ne nf ke ng nh kg ni nj id nk nl ig nm nn ij no cx\">17. Provide global controls.<\/h4>\n<h4>18. Notify users about changes.<\/h4>\n<p id=\"ac4b\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\">The guidelines in this group remind us that AI systems are like getting a new puppy: they are long-term investments and need careful planning so they can learn and improve over time. Learning (Guideline 13) also means that AI systems change over time. Changes need to be managed cautiously so the system doesn\u2019t become unpredictable (Guideline 14). You can help users manage inherent consistencies in system behavior by notifying them about changes (Guideline 18).<\/p>\n<p id=\"b0af\" class=\"ka kb bj kc b hv kd ke hy kf kg kh ki id kj kk ig kl km ij kn gv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">Ideas in Excel<\/strong>\u00a0empowers users to understand their data through high-level visual summaries, trends, and patterns. It\u00a0<em class=\"ko\">encourages granular feedback\u00a0<\/em>(Guideline 15) on each suggestion by asking, \u201cIs this helpful?\u201d<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter wp-image-677478 size-full\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-4.png\" alt=\"Screenshot of Auto Alt Text. Text at top reads Example in Practice\" width=\"406\" height=\"576\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-4.png 406w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2019\/03\/Guidelines_image-4-211x300.png 211w\" sizes=\"auto, (max-width: 406px) 100vw, 406px\" \/><\/p>\n<p id=\"fe3e\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\"><strong class=\"kc mw\">We\u2019d love to hear about your experiences with the guidelines, so tweet us your thoughts <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"http:\/\/www.x.com\/MicrosoftRI\">@MicrosoftRI<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> or like us <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"http:\/\/www.facebook.com\/MicrosoftRI\">on Facebook<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> and join the conversation.<\/strong><\/p>\n<p id=\"5096\" class=\"ka kb bj kc b hv mf ke hy mg kg kh mh id kj mi ig kl mj ij kn gv cx\" data-selectable-paragraph=\"\"><em>Mihaela Vorvoreanu\u00a0is a program manager working on human-AI interaction at Microsoft Research. Saleema Amershi\u00a0is a researcher working on human-AI interaction at Microsoft Research AI. Penny Collisson\u00a0is a user research manager working on AI in Office.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>These 18 guidelines can help you design AI systems and features that are more human-centered. This comprehensive list is based upon more than two decades worth of research across the design community and can be applied to a wide range of products. <\/p>\n","protected":false},"author":39057,"featured_media":677502,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":616842,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[],"msr-locale":[268875],"msr-post-option":[],"class_list":["post-677448","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-en_us"],"msr_assoc_parent":{"id":616842,"type":"group"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/677448","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/39057"}],"version-history":[{"count":13,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/677448\/revisions"}],"predecessor-version":[{"id":677511,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/677448\/revisions\/677511"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/677502"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=677448"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=677448"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=677448"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=677448"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}