{"id":702445,"date":"2020-11-10T08:00:09","date_gmt":"2020-11-10T16:00:09","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-blog-post&p=702445"},"modified":"2020-11-10T09:37:30","modified_gmt":"2020-11-10T17:37:30","slug":"want-to-build-trust-in-your-ai-here-are-3-mistakes-to-avoid","status":"publish","type":"msr-blog-post","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/articles\/want-to-build-trust-in-your-ai-here-are-3-mistakes-to-avoid\/","title":{"rendered":"Want to build trust in your AI? Here are 3 mistakes to avoid"},"content":{"rendered":"

By Penny Collisson (opens in new tab)<\/span><\/a><\/p>\n

\"A<\/p>\n

 <\/p>\n

No longer the stuff of science fiction,\u00a0<\/span>artificial intelligence (<\/span>AI<\/span>)<\/span>\u00a0<\/span>is everywhere today. But, if\u00a0<\/span>we want\u00a0<\/span>AI\u00a0<\/span>to really benefit people, we need to find a way to get them to trust it<\/span>, because people\u00a0<\/span>won\u2019t<\/span>\u00a0use technology they\u00a0<\/span>d<\/span>on\u2019t trust<\/span>.<\/span>\u00a0<\/span>D<\/span>espite<\/span>\u00a0growing advances in AI tech, people still deeply lack confidence in its predictions. Given the option to trust a human over a machine,\u00a0<\/span>they\u2019ll trust a human time and time again<\/span> (opens in new tab)<\/span><\/a>, even if\u00a0<\/span>the human is<\/span>\u00a0clearly wrong<\/span> (opens in new tab)<\/span><\/a>.\u00a0<\/span>\u00a0<\/span><\/p>\n

This is in\u00a0<\/span>part because AI is often working behind the scenes,\u00a0<\/span>so\u00a0<\/span>many people simply aren\u2019t tuned into the times it works. Instead, the moments it goes woefully wrong leave a lasting impression:\u00a0<\/span>a<\/span>n upstanding\u00a0<\/span>citizen being misidentified as a criminal<\/span>, or\u00a0<\/span>items being mistakenly ordered by smart speaker<\/span>s.<\/span>\u00a0<\/span><\/p>\n

We have our work cut out for us if we want to deliver AI in a way that leaves people trusting it.\u00a0<\/span>Some\u00a0<\/span>answers\u00a0<\/span>have come from<\/span>\u00a0recent research<\/span>*<\/span>, which is helping to design intelligence that will integrate seamlessly into people\u2019s lives.\u00a0<\/span>\u00a0<\/span><\/p>\n

\"chart<\/span><\/p>\n

The trust curve<\/span><\/i>\u00a0<\/span><\/p>\n

The\u202f<\/span>trust curve<\/span> (opens in new tab)<\/span><\/a>\u202f(left) highlights three important stages of trust development. At any one of these stages it is possible to either retain or lose users due to a breach of trust: 1. Before use, 2. First moments, 3. Repeat use<\/span>.<\/span>\u00a0<\/span><\/p>\n

S<\/span>imilarly, there<\/span>\u00a0<\/span>are\u00a0<\/span>opportunities to \u2018do it right\u2019 at each stage:\u00a0<\/span>\u00a0<\/span><\/p>\n

    \n
  1. Before use, set positive and\u00a0<\/span>accurate expectations.\u00a0<\/span>People evaluate experiences relative to\u00a0<\/span>what they anticipate<\/span>. Expectations can be positive or negative; accurate or not. If expectations are too low, people won\u2019t engage. If expectations are too high, people will<\/span>\u00a0quickly<\/span>\u00a0feel disappointed and lose trust.<\/span><\/span><\/li>\n
  2. During the\u00a0<\/span>first moments<\/span>\u00a0of\u00a0<\/span>the\u00a0<\/span>experience<\/span>,\u00a0<\/span>set a high bar for value and relevance.\u00a0<\/span>These f<\/span>irst moments are\u00a0<\/span>critical<\/span>\u00a0for building trust. If the experience isn\u2019t immediately seen as\u00a0<\/span>valuable, people won\u2019t use it. If a trust breach occurs early,\u00a0<\/span>curiosity and trust drop<\/span>\u00a0off fast, and people are very unlikely to use the experience again.<\/span>\u00a0Start where your AI is more likely to be right<\/span>:<\/span>\u00a0<\/span>Algorithms aren\u2019t perfect, and there are places where mistakes are more critical than others.\u00a0<\/span>While the AI is learning, it will make mistakes, perhaps doing things the user doesn\u2019t expect. So don\u2019t try to program high value tasks at this stage.\u00a0<\/span>Once th<\/span>ere\u2019s some trust establish<\/span>ed<\/span>,<\/span>\u00a0<\/span>take risks when the user\u2019s relationship with it is farther along.<\/span>\u00a0<\/span>
    \n<\/span><\/li>\n
  3. Reward repeat usage by making the experience more personal over time.\u00a0<\/span><\/span>The simple act of using something makes it feel more familiar, and\u00a0<\/span>breeds<\/span>\u00a0trust. Once you get people engaging beyond the first moments, the next step is to keep them investing.<\/span><\/span>\u00a0<\/span><\/li>\n<\/ol>\n

    What\u2019s\u00a0next<\/span><\/i>\u00a0<\/span><\/p>\n

    We\u2019re making an ambitious\u00a0<\/span>promise with AI. We want it to be personal, contextual,\u00a0<\/span>anticipatory, self-adapting, and\u00a0<\/span>compelling. We want to strike<\/span>\u00a0the right balance between using data\u00a0<\/span>that<\/span>\u00a0create<\/span>s<\/span>\u00a0intelligent, personal experiences\u00a0<\/span>while maintaining customers\u2019 security and privacy.\u00a0<\/span>\u00a0<\/span><\/p>\n

    At Microsoft, we\u00a0<\/span>have a centralized approach to\u00a0<\/span>responsible AI, led by Microsoft\u2019s AI, Ethics, and Effects in Engineering and Research (AETHER) Committee and its working groups, along with our Office of Responsible AI (ORA). Together,\u00a0<\/span>Aether<\/span>\u00a0and ORA work closely with our responsible AI advocates and teams to uphold Microsoft responsible AI principles in their day-to-day work.<\/span>\u00a0Read\u00a0more\u00a0<\/span>here<\/span><\/a>.<\/span>\u00a0<\/span><\/p>\n

    As\u00a0<\/span>with any relationship, it all starts with trust.\u00a0<\/span>\u00a0<\/span><\/p>\n

    * There were three phases of research used to glean these findings:\u00a0<\/span>\u00a0<\/span><\/p>\n