10/7 THE RECURRENT SITUATION 



conditions to which it was already adapted at D x ; yet after the 

 events following the first reversal (at R^), the first adaptation (at 

 Dj) is totally lost; and the Homeostat treats the situation after 

 R 2 as if the situation had occurred for the first time. 



In general, if the Homeostat is given a problem A, then a prob- 

 lem B, and then A again, it treats A as if it had never encountered 

 A before; the activities during the adaptation to B have totally 

 destroyed the previous adaptation to A. (The psychologist would 

 say that retroactive inhibition was complete, S. 16/12.) 



This way of adapting to A on its second presentation cannot be 

 improved upon if the environment is such that there is no implica- 

 tion that the second reaction to A should be the same as the first. 

 The Homeostat's behaviour might then be described as that of a 

 system that ' does not jump to conclusions ' and that ' treats every 

 new situation on its merits '. In a world in which pre-baiting was 

 the rule, the Homeostat would be better than the rat ! When, 

 however, the environment does show the constraint assumed in 

 this chapter, the Homeostat fails to take advantage of it. How 

 should it be modified to make this possible ? 



10/6. The Homeostat has, in fact, a small resource for dealing 

 with recurrent situations, but the method is of small practical use. 

 In S. 8/10 we saw that the Homeostat's ultimate field is one that 

 is stable to all the situations, so that a change from one to another 

 demands no new trials. 



10/7. This method, however, cannot be used extensively in the 

 adaptations of real life, for two reasons. The first is that when 

 the number of values is increased beyond a few, the time taken 

 for a suitable set of step-function values to be found is likely to 

 increase beyond anything ordinarily available, a topic that will 

 be treated more thoroughly in Chapter 11. The second is that 

 the adaptation, even if established, is secure only if the set of 

 parameter-values is closed, i.e. so long as no new value occurs. 

 Should a new value occur, everything goes back into the melting- 

 pot, and adaptation to the new set of values (the old set increased 

 by one new member) has to start from scratch. Common observa- 

 tion shows, of course, that each new adaptation does not destroy 

 all the old; evidently the method of S. 8/10 is of little practical 

 importance. 



141 



