COMPARING THE BRAIN WITH MACHINES — MacKAY 233 



The reason is that by a simple logical process we can turn any such 

 precise test into a description of at least one mechanism that will meet 

 it. It may be an inordinately bulky mechanism, but we shall be unable 

 to distinguish its performance from the behavior that has been 

 specified. 



It follows that all arguments that begin : "You'll never get a ma- 

 chine to do such-and-such," are foredoomed as soon as the speaker has 

 been induced to say precisely what behavior he would regard as 

 satisfactory. What is doubtful, of course, is whether we shall ever be 

 clever enough to specify, even in principle, an exhaustive test for 

 humanlike behavior. It is this aspect of the problem, rather than the 

 mechanical difficulty, that I believe is of more than trivial philosoph- 

 ical interest (Wisdom et al., 1952) . 



The third question I have raised is actually the one of greatest prac- 

 tical importance : How far is it possible to envisage an artificial mecha- 

 nism that would not only imitate human behavior, but work internally 

 on the same principles as the brain ? It is this possibility that I want 

 chiefly to discuss. How do we set about comparing the brain with ma- 

 chines in this sense? How far have we progressed, how far can we 

 hope to go — and what is the point of it all anyway ? 



METHODS OF APPROACH 



The brain is unique in offering two kinds of clues as to its function, 

 in physical observation on the one hand and psychological study on the 

 other. Each of these methods requires a very different language, and 

 the problem is to find a common language in which to associate the two. 

 Fortunately, the language of information and control theory turns out 

 to be just what is needed, since it belongs in a sense to both fields. 



The first step, therefore, in designing a theoretical model must be to 

 sum up what we know of the way in which the brain handles informa- 

 tion, at all levels from the nerve cell up to the complete organism. This 

 provides less positive guidance than one might think, but it does set 

 certain negative limits on the kind of model we may envisage (see, for 

 example, Grey-Walter, 1953). 



We next ask : Within these limits, what mechanisms are there that 

 could handle and respond to information in the same way ? Again we 

 can tackle this question at various levels. The exponents of what is 

 called "nerve-net theory" start with an idealized model of the nerve 

 cell, and develop from it theoretical networks capable of complex func- 

 tions.^ My own preference is for a statistical model of the whole 

 system, with a flexible structure that may be brought gradually to re- 

 semble the large-scale structure of the nervous, and humoral, system 

 (MacKay,1952). - . 



' Most of this work has appeared in Bull. Math. Biophys. from 1943 to date. 



