DESIGN FOR A BRAIN 10/8 



The accumulator of adaptations 



10/8. To see what is necessary, let us take for granted that 

 organisms are usually able to add new adaptations without destroy- 

 ing the old. Let us take this as given, and deduce what modifica- 

 tions it enforces on the formulation of S. 7/5. Suppose, then, 

 that an organism has adapted to a value P 1% has then adapted to 

 P 2 by trial and error as in S. 7/23, and that when P 1 is restored 

 the organism is found to be adapted at once, without further 

 trials. What can we deduce ? 



(The arguments that culminated in S. 7/8 apply here without 

 alteration, so we can take for granted that the adaptation to each 

 individual value of P takes place through the second feedback, 

 with essential variables controlling step-functions as in S. 7/7. 

 The modification to be made can be found by a direct application 

 of the method of S. 4/12, seeing whether variation at one variable 

 leads to variation at another.) 



To follow the argument through, let us define two sub-sets of 

 the step-mechanisms in S that affect R (Figure 7/5/1): 



S x : those step-mechanisms whose change, with P at P v would 

 cause a loss of the adaptation to P x (i.e. those step- 

 mechanisms that are effective towards R when P is at 



Pi); 



S 2 : those step-mechanisms that were permanently changed in 

 value after the trials that led to the adaptation to P 2 . 



First it follows that the sets S x and S 2 are disjunct, i.e. have no 

 common member. For if there were such a common member it 

 would (as a member of S 2 ) be changed in value when P x was 

 applied for the second time, and therefore (as a member of S x ) 

 would force the behaviour at P x to be changed on P x 's second 

 presentation, contrary to hypothesis. Thus, for the retention of 

 adaptation to P lf in spite of that to P 2 , the step-mechanisms must 

 fall into separate classes. 



(That the step-mechanisms must be split into classes can be 

 made plausible by thinking of the step-mechanisms, in any ultra- 

 stable system, as carrying information about how the essential 

 variables have behaved in the past. When P x is presented for the 

 second time, for the behaviour to be at once adaptive, information 

 must be available somewhere about how the essential variables 



142 



