16/12 ADAPTATION IN THE MULTISTABLE SYSTEM 



Before attempting the answer, let us recall that, in any poly- 

 stable system, any two different lines of behaviour will give 

 changes in two sets of variables (S. 13/14) which may or may not 

 overlap. Each set will be distributed over the system somewhat 

 as the smoking chimneys of S. 13/18 were distributed over the 

 town. Two disturbances (D 1 and D 2 ), to a polystable system, 

 will give two sets of active variables, as two winds (W 1 and W 2 ), 

 to a town, would ?give two sets of smoking chimneys. 



Of the chimneys in the town, what fraction will smoke with 

 both the winds ? The precise answer would depend on precise 

 conditions; but we can see as a first approximation (as in S. 13/15) 

 that if only a small fraction smoke under W v and a small fraction 

 under W 2 , then if the two fractions are independent, the fraction 

 smoking under both will be the product of the single fractions, and 

 thus much smaller than either. Thus if a random 1 per cent 

 smoke under W 1 and another random 1 per cent under W 2 , those 

 that smoke under both will be only y^- of 1 per cent. 



The independence, and the smallness of the overlap, can occur 

 only if W x and W 2 are well separated in direction. If W 2 should 

 be very close to W^s direction, it will probably cause smoking in 

 many of I^'s chimneys (in the limit, of course, as it matches 

 W^s direction, it will make all W^s chimneys smoke). 



Thus a polystable system (subject to certain conditions of 

 statistical independence which would require detailed examina- 

 tion) will respond to two parameter-values (or disturbances, or 

 stimuli) with two sets of variables whose overlap depends on: — 

 (1) the amount of activation that each causes, and (2) the resem- 

 blance between the parameter- values. 



Suppose now that the parameter values correspond, as in 

 S. 10/8, to environments that have to be adapted to (or to problems 

 that have to be solved). Since the multistable system is also 

 polystable, what has just been said will be true of the multistable 

 system. Here the two lines of behaviour will include trials and 

 will cause changes in the step-mechanisms as well as in the main 

 variables. The degree to which the two sets of activated step- 

 mechanisms overlap will again depend on what fraction of all 

 step-mechanisms are activated and on the degree of resemblance 

 of the parameter- values (or environments). In particular, if the 

 lines of behaviour overlap on only a few step-mechanisms, the 

 second set of trials may cause little change in the step-mechanisms 



215 



