9/5 DESIGN FOR A BRAIN 



Once it is agreed that a system, such as that of Mowrer's rat, 

 contains step-functions, then all it needs is that they should 

 not be few for the system to be admitted as ultrastable. 



After this, we can examine the qualifications that were added 

 when considering Stentor as an ultrastable system. Are they, 

 too, necessary ? Not with the assumptions made so far in this 

 section, but they become so if we add the postulates that the 

 system ' adapts ' in the sense of S. 5/8, and that it does so by 

 ' trial and error'. In order to be definite about what 'trial and 

 error ' implies, here is the concept defined explicitly : 



(1) The organism makes trials only when ' dissatisfied ' or 



4 irritated ' in some way. 



(2) Each trial persists for a finite time. 



(3) While the irritation continues, the succession of trials 



continues. 



(4) The succeeding trial is not specially related to the preced- 



ing, nor better than it, but only different. 



(5) The process stops at the first trial that relieves the irritation. 

 The argument goes thus. As each step-function forms part 



of an absolute system, its change must depend on its own and 

 on the other variables' values ; there must, therefore, be certain 

 states — the critical — at which it changes value. When, in the 

 process of adaptation by trial and error, the step-function changes 

 value, its critical states must have been encountered ; and since, 

 by (1) above, the step-functions change value only when the 

 organism is ' dissatisfied ' or ' irritated ', the critical states must 

 be so related to the essential variables that only when the organism 

 is driven from its normal physiological state does its representa- 

 tive point encounter the critical states. This knowledge is suffi- 

 cient to place the critical states in the functional sense : they 

 must have values intermediate between those of the normal state 

 and those of the essential variables' limits. The qualifications 

 introduced in S. 9/1 are thus necessary. 



Training 



9/5. The process of ' training ' will now be shown in its relation 

 to ultrastability. 



All training involves some use of ' punishment ' or ' reward ', 

 and we must translate these concepts into our form. ' Punish- 



112 



