Just to add two comments to Jon's posting:
Jon Williamson wrote: > > Practically every science faces the difficult task of sharpening qualitative > perceptual judgements into the precise, quantitative language of the science > in question. The way a science does this is rarely formulated explicitly as > part of the science itself, and often seems mysterious from the outside. > Probability theory, for instance, tells you what to do when you have certain > probabilities and certain assumptions hold, not how to arrive at this > information. J. M. Keynes thought that reasonable people would agree on the assignment of probabilities. Clearly, he didn't know many reasonable people! >It is the job of statisticians, knowledge engineers and > philosophers of science to better articulate the sharpening process, but > just because it isn't written down in books on probability theory, that > doesn't mean it can't be done. > I believe there is a key difference in objectives here between statisticians, on the one side, and artificial intelligence researchers, on the other. AI needs to articulate computational processes by which probabilities (or the values obtained using other uncertainty formalisms) can be assigned, so that machines can be built which can assign probabilities using these processes. As with the processes underlying any software system, these processes need to be shown to have desirable properties before deployment. For statisticians, especially those who view their role as an advisory one to a decision-maker, it is enough that probabilities can be assigned. Indeed, we have a whole branch of statistics which assumes that probabilities arise entirely subjectively, ignoring completely the question of how this happens. Thus, despite a public claim I once heard made by a well-known statistician, AI is not a sub-branch of statistics. -- Peter McBurney University of Liverpool