Tuesday, September 14, 2010
Do we know what we don't know?
How good are we in estimating the uncertainty of our claims? Pretty bad in my opinion. And this may be particularly true for scientists, medical doctors or other experts. Since they have noticed that there are very few people who know more than them on a (very) particular topic they infer that they may actually know close to everything there is to know about it. I recently watched in interview with a renowned physicist about the possibility of EPS (extra perceptual sensation). This expert embarked on a long story about how all physical laws decay as one over the square of distance and that therefore the signals necessarily underlying ESP would have been detected. Problem is of course that his reasoning was solidly rooted in the physical laws as we know them and that the possibility of entirely new physics causing the phenomenon was simply denied. A clear overestimation of ones grasp of the unknown.
Ever tried to argue with a doctor why your child needs Tylenol for a mild fever? They will almost consider you criminal if you choose to deny them the medicine. But there is never a clear reason as to why they need it. It's simply the way it is. But do they have any clue as to the long term effects of poring medicine into these small bodies? Yes, of course, Tylenol was rigorously tested and approved but it's almost impossible to test for the increased risk of cancer after 20 years. They seem completely certain it's safe until a new study shows it's not (as was indeed the case with Tylenol). Why are doctors so certain about the effect of drugs or vaccinations: because they chronically overestimate their grasp of what is unknown.
Perhaps the clearest example is given by reviews of scientific papers where reviewers are asked to provide their confidence. It is very common to find two maximally confident reviewers with completely opposite opinions. Clearly one of the two must be wrong. And, this is of course very frustrating at the receiving end.
Bottom line, always keep an open mind and try not to underestimate what you don't know.
Subscribe to:
Post Comments (Atom)
Though I fully agree, I'm not sure whether the doctor example is a very good example.
ReplyDeleteI did a project once for which I had to read parts of Harrison's internal medicine Bible. What struck me was the high incidence of the word "unknown": virtually every article described phenomenons, characteristics, or treatments that are poorly understood. Assuming that doctors studied this book (and many others), they should thus be very well aware of the stuff they don't know or understand.
I think the prescription of Tylenol (and a lot of other useless treatments) is more likely to be the result of the pressure on doctors from the outside world. Such pressure comes from pharmaceutical companies, who invite doctors to "study weekends" about their new medicines (the word bribe is not completely inappropriate here). But it also comes from patients, who aren't exactly satisfied if their doctor (1) tells them he doesn't know and/or (2) does not prescribe any treatment.
Having said that, there is one thing that doctors don't generally know about even though they should: Bayes' rule!
One of the best college courses I ever took was "The History of Science." This showed simple examples like the grapes in jello model of the atom being the accepted model that everyone agreed they were sure was right. Until they were wrong. and the obvious example of Galileo who discovered "new" truth that was not accepted because it could not possibly be correct according to the current thinking.
ReplyDeleteAll scientists and scholars should note this well. Perhaps better technology will allow us to better observe what we cannot currently.
I agree that to a proper Bayesian, absolutely certainty does not exist. On the other hand, a common cognitive effect is for people to overestimate small probabilities and see meaning in coincidences. People who believe in ESP are an example of that. While one cannot prove that ESP is impossible, I think it is appropriate for a scientist to point out its very small probability given available evidence.
ReplyDelete