Moravec’s paradox and the rash

by reestheskin on 09/11/2016

Comments are disabled

In all places, I came across the following in Dylan Wiliam’s most recent book (if you want to understand what teaching feedback really is, read it). After pointing out how some machine learning techniques can outperform some medics in some contexts, he writes

However, it is important to realize that the key factor in making jobs suitable for automating is not that they are manual or low skill. It is that they are routine. A task can require many years of training for humans to become good at it, but it can still be relatively routine, thus making it relatively straightforward to automate. This is just one example of a much more general principle, which is that many of the things that we thought would be easy to automate turn out to be rather complex, while many of the things that we thought would be hard to automate turn out to be reasonably simple. …. This observation—that high-level reasoning seems to require very little in the way of machine power, while many low-level sensorimotor skills appear to require huge computational resources—is known as Moravec’s paradox, named after Hans Moravec, who pointed out that “it is comparatively easy to make computers exhibit adult-level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility” (Moravec, 1988, p. 15).

Now, of course I work in a domain that is heavily perceptual and, as yet, AI systems have made little inroad. This may well be because the task is difficult (and that the human visual system is powerful) but also (critically) that the available training sets are orders of magnitude too small. This will only change if the clinical workflow is fully digital. We have published some work in this area, and if you visit the Dermofit app on the iOS store you will see an app that uses some machine learning. But a long, long way to go, and the humans can still pay the bills. For the moment.

The question for medical education is that as we (reasonably) concentrate our procedures around high level processing, the sort of environments we need to develop perceptual skills are neglected. You can do both.