How do you come to terms with all those algorithms that increasingly pervade our lives? How do you know you can trust them? Turns out that the answer is in human psychology and our need for a Goldilocks explanation.
Some algorithms merely make our lives easier. Amazon recommends new or additional purchases. Netflix suggest movies we might like. Match.com presents people we might want to date. Some algorithms are not so benevolent. Judicial sentencing recommendations rank black defendants as higher risk than whites. Mortgages are granted upon the zip code you live in. Applying for a job? An algorithm will review your application first. Don’t be the wrong age; you’ll never get an interview.
Algorithms embedded in Artificial Intelligence and Machine Learning are increasingly coming to medicine. How are you going to feel when the doctor (if you even get to see one) informs you that “the computer” ranks your risk of disease at above average and you should get your affairs in order? If you are like most of us, I think you would be digging into the “why” of “the computer’s” rationale.
What level of detail in an explanation would satisfy you? Surprisingly, a researcher at Stanford, René Kizilcec, discovered that we anthropomorphize (project human characteristics) computers. We feel they are “thinking” – sort of like us. Don’t believe it? Consider how you feel when your laptop will not do what you want it to. Do you feel that it doesn’t like you or maybe it’s being malicious?
Now, think about how you react when you meet a person you know nothing, or very little, about. You are at first cautious and your trust level is low. So you follow certain social protocols of introduction: Nice to meet you. How are you? Where are you from? What do you do? Gradually, you build up a level of understanding; trust and comfort with the stranger. You’ll go on to see if you have commonalities in your backgrounds and interests – so far, so good.
But what happens if the stranger, now your new found acquaintance, keeps going and begins to over share. We have all been there: that moment of discomfort when you are getting way too much detail about personal or private issues. Pretty soon you are trying to figure out a way to disengage and get out of the conversation.
For human beings there is such a thing as the “right” amount of transparency – not too little, not too much. What’s this got to do with algorithms? Kizilcec argues that the same applies to algorithms. Too much information can undermine user trust as much as too little – aka Goldilocks. Kizilcec believes from her research that there is a similar sweet spot for trust in human-algorithm interactions.
We need just enough explanation to make us feel that the process is fair. Beyond that our trust level declines, especially if the outcome was disappointing.
This post is based upon an article in Wired: People Want to Know About Algorithms—but Not Too Much and a new book: A Human’s Guide to Machine Intelligence both by Kartik Hosanagar.