DESIGN FOR A BRAIN 17/11 



Distribution of feedback 



17/11. Another adjustment that is necessary, if the adaptation 

 is to be more than merely nominal, has already been made in 

 Figure 16/6/1, which thereby begged an important question. In 

 the Figure, if we start in the environment at any subsystem and 

 trace a route through the essential variable that it affects, on 

 through the corresponding step-mechanism, reacting part, and so 

 back to the environment, we arrive at the same subsystem as the 

 one we started at. The Figure thus implies that if an essential 

 variable, E x say, is being upset by a part of the environment, 

 E^s actions will eventually affect the very part of the environment 

 that is the cause of the trouble. 



The correspondence undoubtedly favours efficiency in adapta- 

 tion, as may be seen by tracing explicitly what would happen 

 otherwise. (The argument is clearest when the systems are 

 iterated, Figure 15/7/1.) Suppose, in it, that the second-order 

 loops were severed and then re-connected in some random way: 

 so that the essential variable of A affected the reacting part of J?, 

 say. A disturbance to A that A is not adapted to would now 

 result in changes at B's step-mechanisms, though the set of values 

 here might be perfectly adapted to dealing with whatever disturb- 

 ance came to B. Thus without proper distribution of the second- 

 order feedbacks the effects from the essential variables would 

 only change at random, destroying in the process minor adapta- 

 tions already established. Thus without appropriate distribution 

 of the second-order feedbacks there cannot be that conservation 

 of correct adaptations in the subsystems, and the cumulative 

 progression to adaptation that Chapter 10 treated as of major 

 importance. The system would still adapt as a Homeostat does, 

 but it would take the excessive time of T 1 rather than the moderate 

 time of T 3 (S. 11/5). 



The distribution of second-order feedbacks cannot be settled 

 once for all, for a part of each circuit is determined by, or supplied 

 by, the environment, and is thus subject to change. To this the 

 organism must make counter-adjustments, if the distribution is 

 to remain appropriate. 



A well-known example that illustrates the necessity for finding 

 where to apply a correction is given by the aspiring chess-player 

 who has just -lost a game and who is considering how his strategy 



228 



