We will see much more automated advice in the years ahead in a whole range of areas, all from financial services to health and medicine. A key question in all of these is how to achieve trust. Perhaps we double check medical conditions but by and large we trust doctors with our health. Maybe we trust financial advice considerably less, but many people still find it comforting to speak to a real person. A robot psychiatrist might work far into the future, but for now humans are still preferred in many service areas.
So what steps do programmers need to consider to improve the trust we place in automated advice?
Having worked in a central bank for many years, I have seen the distrust with which forecasts from technical models are viewed. I believe the mechanism with robo-advice is the same. Black boxes, no matter how sophisticated, are viewed with suspicion. Advice that we do not understand is discounted, even if phrased in simple words.
The other day I was driving my car to work and the iPhone prompted me that there was unusually heavy traffic. That is vague enough to satisfy even the most suspect astrology forecast. But in addition, it also specified that the trip would take 1 hr and 13 minutes, about half an hour longer than usual. This was a wonderfully precise estimate. But I did not believe it!
Let's come back to accuracy of that number in a few moments. The reason I did not trust it was that it was such a large deviation from my expectation; also, the combination with the "unusually heavy traffic" seemed too generic. What information was it based on?
I think for me to trust the time estimate, the program should have given me some reasons for the heavy congestions. For example, an accident on road x, a trade show in town or even pictures of slow moving traffic. Any kind of information that substantiated the claim.
Had I known about the traffic situation, I would have parked my car before Stockholm and taken the bus instead.
And, you guessed it, the time estimate was fairly good! But I still do not know if it was a lucky guess or based on solid method! Maybe try again tomorrow...