REGULATING THE VERY LARGE SYSTEM 13/2 



correlated with the variety in D for several reasons. If T is made 

 of many parts, and there is uncertainty about the initial state of any 

 part, then that variety will be allocated to D (S.l 1/19); so in general, 

 other things being equal, the greater the number of parts the greater 

 the variety in D. Secondly, if each part is not completely isolated 

 from the world around, each part's input will contribute some variety 

 which will be allocated to D; so in general, the greater the number of 

 parts the greater the number of components in D; and therefore, 

 if the components have some independence, the greater the variety 

 in D. (There may be other reasons as well but these will suffice.) 



Thus, when the effects of size are distinguished from those that 

 affect the variety in D, it will usually be found that the former is, 

 in itself, irrelevant, and that what matters is the latter. 



It now follows that when the system T is very large and the regu- 

 lator R very much smaller (a common case in biology), the law of 

 Requisite Variety is hkely to play a dominating part. Its importance 

 is that, if R is fixed in its channel capacity, the law places an absolute 

 limit to the amount of regulation (or control) that can be achieved 

 by R, no matter how R is re-arranged internally, or how great the 

 opportunity in T. Thus the ecologist, if his capacity as a channel is 

 unchangeable, may be able at best only to achieve a fraction of what 

 he would hke to do. This fraction may be disposed in various ways 

 — he may decide to control outbreaks rather than extensions, or 

 virus infections rather than bacillary — but the quantity of control 

 that he can exert is still bounded. So too the economist may have 

 to decide to what aspect he shall devote his powers, and the psycho- 

 therapist may have to decide what symptoms shall be neglected and 

 what controlled. 



The change in the point of view suggested here is not unUke that 

 introduced into statistics by the work of Sir Ronald Fisher. Before 

 him, it was taken for granted that, however clever the statistician, a 

 cleverer could get more information out of the data. Then he 

 showed that any given extraction of information had a maximum, 

 and that the statistician's duty was simply to get near the maximum — 

 beyond that no man could go. Similarly, before Shannon's work it 

 was thought that any channel, with a little more skill, could be 

 modified to carry a little more information. He showed that the 

 engineer's duty is to get reasonably near the maximum, for beyond 

 it no-one can go. The law of Requisite Variety enforces a similar 

 strategy on the would-be regulator and controller: he should try 

 to get near his maximum — ^beyond that he cannot go. Let us 

 therefore approach the very large system with no extravagant ideas 

 of what is achievable. 



245 



