Although genetics once entranced me, I take the view that the subject’s impact on medicine (as compared with biology) will be more modest than many imagine. Genetics is a wonderful way to do biology but biology is not synonymous with medicine: getting this relation wrong is one of the reason much medical research is so dysfunctional (see an article I wrote in Science in 2002).
To quote Roger Needham, ‘It is easy to do research if you think research is finding out what nobody know. Well that isn’t good enough; if you want to do research, you want to do research that will have some influence’.
In the areas I know about, there had been a marked dissociation between real clinical advance and much professional career research, and in my own case, although I had had a good and fun run with the genetics of red hair, I worried I would end up on the wrong side: I needed to do something different.
At the same time as I was having these doubts (see my article in Science in 2002) I chanced upon the obituary of the Nobel Laureate Herbert Simon in Science. I had never heard of him before and I subsequently picked up a copy of ‘Sciences of the Artificial’ and read some of his other work. It was an epiphany and I have never been able to see medicine or medical research in the way I used to. Quoting Roger Needham again, ‘When I was an academic I used to interview candidates for professorships at other universities and I always asked them what they thought their research activities would be in the next ten years, and the only correct answer was, ‘I don’t know’’.
I didn’t and still don’t, but the gedanken experiments that fascinate me are:
- What information would it take to build a dermatologist?
- How can we simulate clinical skills and improve upon them?
- How do we attach semantics to images?
- Can we invent cognitive prostheses that improve clinical skills?
- How can we use machines to teach humans how to diagnose?
I think we will only make real progress when we build machines to rival, augment or supplant our own skills. Medicine is therefore a branch of engineering: we build systems on the basis of our knowledge of the natural world. This is dermagineering and is what I do.
The central skill of a dermatologist is in the recognition of morphology. This of course is not all there is to clinical practise but just as for the radiologist or the pathologist, the ability to recognise form is critical. The research problem —and it is non-trivial — is how to attach semantics to images. Books have word indexes, as does the Web, but to find out more about a rash you have to be able to name it first. So, the traditional description (or put-down) of dermatologists as butterfly collectors or stamp collectors (adding petals to their dermatological colouring book as Sam Shuster said) is spot on! But it is an amazing skill and cognate with so many other human skills that we currently find difficult to program machines to do. Just think of the way children learn to distinguish cats from dogs, or the way humans can identify faces or face-like objects in a complex scene.
What continues to amaze me is just how informative these skills are of the world we live in. It seems we are able to bootstrap so much of the world purely on the basis of classification of form and what has happened in the past. So, without any knowledge of pigmentation biology, or of physiology, we can associate the particular surface morphologies of a cheetah, leopard or panther and work out whether it is better to avoid eye contact with the cat or not. To give another example, pathologists without knowing anything about DNA, chromosomes, or nuclear structure were able to link variation in nuclear shape with biological behaviour. Even today in dermatology, a simple H & E stain provides more information than sequencing any number of cancer associated genes. The Reverend Bayes would be impressed.
Now there are those dinosaurs amongst my colleagues who think it is better to leave well alone in areas like this. The argument goes that only humans possess such visual skills and that it will be beyond machines. May be, but I will bet against this view. Doctors used to taste urine to diagnose diabetes: fortunately I am not that old. Remember, physicians used to claim that you could measure the blood pressure with a skilled finger on the pulse. Well, the boundaries between man and our chimp cousins may be increasingly blurred but humans remain unmatched as tool makers. Acquiring clinical skills is expensive and for much of the world there are not enough trained practitioners. In reality the average clinical abilities of those called upon to diagnose skin diseases even in wealthy Western Europe is far from ideal.
There is another argument in favour of automation that I owe to Frank Zappa. When asked why he didn’t tour any more or work with other musicians, instead relying on an early electronic composing tool, he replied that rock musicians are a troublesome bunch, requiring to be paid and often ending up getting busted at customs. Machines have their attractions.
For exactly what we doing look under Dermofit.