232 ANNUAL REPORT SMITHSONIAN INSTITUTION, 1954 



suggested that a probabilistic model, capable of altering its statistical 

 rules of activity according to experience, offers good prospects of suc- 

 cess as a research tool. 



A good model is judged not only by its predictive power, but by the 

 clarity with which its failures offer constructive interpretations. Un- 

 derstanding is to be distinguished from ability to predict. Even a 

 complete mathematical equation for the nervous system would con- 

 tribute little to the understanding we require, and the useful role of 

 conventional mathematics is likely to be small. 



Perhaps the major aim of this work is to provide a working bridge 

 between the languages of psychiatry and physiology, for the better 

 coordination of attacks on mental disease. 



Such studies are sometimes attacked as "debimking man." This 

 appears to be a misconception. I shall suggest (a) that ethical re- 

 sponsibility in no way hangs on physical indeterminacy in the brain, 

 and (h) that physical description and mental description reveal 

 complementary aspects of one and the same human activity, which in 

 its full nature combines and transcends both. 



COMPARING THE BRAIN WITH MACHINES 



Wlien we speak of comparing the brain with machines we may mean 

 one of three things, according to which one of three quite different 

 questions we may have in mind. 



We may want to know to what extent the brain resembles existing 

 mechanisms, such as electronic computing machines. The short an- 

 swer to this question is that electronic computers are not designed to 

 resemble the brain : in fact they are explicitly designed not to resem- 

 ble the brain in most of its important functions. A computer that 

 developed esthetic preferences and had occasional ideas of its own 

 would have a limited field of usefulness — or at least an unusual one ! 

 Any resemblances there may be are for our present purpose trivial. 



The second question we might have in mind is whether an artificial 

 mechanism, or "artifact," could ever be designed to imitate all human 

 behavior. Could such an artifact frame novel hypotheses or write son- 

 nets, for example ? 



This question is slightly more interesting from some points of view. 

 By imitation, of course, we should mean much more than the sort of 

 imitation provided by a phonograph or even a walking, talking robot. 

 In fact, we are entitled to demand that the artifact would have to meet 

 every functional test which the brain — or the nervous system as a 

 whole — can meet. 



But the answer to this question also is fairly straightforward. Any 

 test that we can specify exactly, or even statistically, in terms of the 

 behavior to be expected in given circumstances, can in principle be met 

 by an artifact built only of mechanisms that are now available to us. 



