{"id":710068,"date":"2020-12-08T09:54:48","date_gmt":"2020-12-08T17:54:48","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&p=710068"},"modified":"2020-12-09T11:33:30","modified_gmt":"2020-12-09T19:33:30","slug":"why-people-may-not-trust-your-ai-and-how-to-fix-it","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/why-people-may-not-trust-your-ai-and-how-to-fix-it\/","title":{"rendered":"Why people may not trust your AI, and how to fix it"},"content":{"rendered":"

By\u00a0<\/span><\/span>Penny Collisson<\/span><\/span><\/span><\/a>,\u00a0<\/span><\/span>Gwen<\/span><\/span>\u00a0Hardiman<\/span><\/span><\/span> (opens in new tab)<\/span><\/a>\u00a0and\u00a0<\/span><\/span>Trish Miner<\/span><\/span><\/span><\/a>\u00a0<\/span><\/p>\n

\"a

Photo credit: iStock<\/p><\/div>\n

Pssst. Users don’t trust your AI. <\/span><\/p>\n

It\u2019s OK. Don\u2019t panic. You have an opportunity to change that<\/span>, and\u00a0here\u2019s\u00a0how.\u00a0<\/span>\u00a0<\/span><\/p>\n

What we know\u00a0<\/span><\/i>\u00a0<\/span><\/p>\n

Despite advances in technology, users are still <\/span>hesitant<\/span>\u00a0to fully trust the capabilities and suggestions of AI.\u00a0<\/span>Given the example of trusting a human or trusting AI, people will choose\u00a0the human,\u00a0<\/span>even when\u00a0the human\u00a0is\u00a0wrong<\/span><\/i>.\u00a0<\/span>Perhaps surprisingly, when the human is shown to be wrong repeatedly<\/span>, the\u00a0<\/span>person seeking suggestions<\/span>\u00a0<\/span>will still choo<\/span>s<\/span>e<\/span>\u00a0the other human.\u00a0<\/span>Here\u2019s\u00a0how to bridge those two worlds and build trust.\u00a0<\/span>\u00a0<\/span><\/p>\n

Designing\u00a0<\/span><\/i>for trust\u00a0<\/span><\/i>\u00a0<\/span><\/p>\n

Designing for trust<\/span><\/a>\u00a0is about putting people at the center of our approach \u2013 their humanity,\u00a0<\/span>intents<\/span>, and feelings \u2013 not technology.\u00a0Empathizing with people, and obsessing\u202f<\/span>over<\/span>\u202four relationship with them, must go hand in hand with\u00a0leveraging\u00a0the superpowers of intelligent\u202ftech. Here are\u00a0<\/span>three\u00a0<\/span>four\u00a0<\/span>practical things you can start doing today to design with trust in mind:\u202f<\/span>\u00a0<\/span><\/p>\n

    \n
  1. Walk<\/span><\/b>\u00a0through\u00a0<\/span><\/b>your\u00a0<\/span><\/b>user\u2019s experience<\/span><\/b>\u00a0as you design<\/span><\/b>.\u00a0<\/span><\/b>S<\/span>ometimes the user\u2019s idea of how a particular feature is supposed to work differs from the designers or the\u202fengineers.\u00a0That\u2019s\u00a0often because people are using your intelligent\u202ffeature\u202fin the context of other things. One trick you can use when evaluating a new\u00a0<\/span>AI<\/span>–<\/span>based\u00a0<\/span>feature is to look at it in the context of a\u00a0likely\u202f<\/span>end\u00a0to end user story.\u00a0<\/span>\u00a0<\/span><\/li>\n
  2. Prototype with <\/span><\/b>real<\/span><\/b>\u00a0user data.<\/span><\/b>\u00a0<\/span>It can be tricky to\u00a0anticipate\u00a0all the ways your AI might go wrong before you have an algorithm. One suggestion is to\u00a0<\/span>walk through your experience yourself using real user content<\/span> (opens in new tab)<\/span><\/a>. You can also\u00a0<\/span>use people\u2019s data<\/span> (opens in new tab)<\/span><\/a>\u00a0to simulate a wrong answer, and then get them to respond. This approach can highlight gaps you might have otherwise overlooked. Once you know the gotchas,\u00a0you\u2019re\u00a0better equipped to make use of the guidance on just how to design for being wrong. Prototyping with\u00a0<\/span>real\u00a0<\/span>data helps you get ahead of being wrong.<\/span>\u00a0<\/span><\/li>\n
  3. Ask people for feedback, then act on\u202fit<\/span><\/b>.\u00a0<\/span><\/b>Research suggests that involving people more with their AI not only improves trust, but also allows the AI to learn from people\u2019s experience of it. This is because it enables users to direct\u00a0AI\u00a0so it\u00a0benefits\u00a0them most and leaves them feeling in control.\u00a0<\/span><\/li>\n
  4. Get the data about\u00a0what\u2019s\u00a0working<\/span><\/b>.\u00a0<\/span><\/b>The next step is to\u00a0<\/span>ensure the experience\u00a0we\u2019re\u00a0shipping, even when it is not\u00a0<\/span>yet<\/span><\/i>\u00a0the best<\/span>\u00a0it can be<\/span>,\u00a0is\u00a0<\/span>good enough<\/span><\/i>\u00a0to get\u00a0<\/span>the\u00a0<\/span>repeat usage<\/span>\u00a0we need to get data about\u00a0what\u2019s\u00a0working<\/span>.<\/span>\u00a0In our work with product teams, we<\/span>\u00a0aim to ship once\u00a0we\u2019re\u00a0confident we have achieved\u00a0a trustworthy\u00a0experience<\/span>.\u00a0<\/span>\u00a0<\/span><\/li>\n<\/ol>\n

    Measuring trust<\/span><\/i>\u00a0<\/span><\/p>\n

    How<\/span>\u00a0do we\u00a0operationalize\u00a0and measure\u00a0<\/span>the trustworthiness of an experience<\/span>\u00a0before we ship?\u00a0<\/span>People\u2019s feelings, thoughts, and actions reflect trust. Looking at all three dimensions helps us get a fuller picture of a human being and\u00a0determine\u00a0whether they trust an experience.<\/span>\u00a0<\/span>Applying the lens of feel-think-act can help you structure your evaluations, make sense of people\u2019s feedback, and justify your decision to move forward and ship, or\u00a0iterate\u00a0one more time.<\/span>\u00a0<\/span><\/p>\n

    \"\"<\/p>\n

    To\u00a0<\/span><\/i>measure\u00a0<\/span><\/i>if an experience is trustworthy, consider how someone will\u00a0<\/span><\/i>feel,\u00a0<\/span><\/i>think,\u00a0<\/span><\/i>and\u00a0<\/span><\/i>act\u00a0<\/span><\/i>\u00a0<\/span><\/p>\n