13/20 AN INTRODUCTION TO CYBERNETICS 



How can a machine select ? The answer must, of course, be given 

 in terms compatible with those already used in this Part. 



Perhaps the simplest process of selection occurs when a machine 

 goes along a particular trajectory, so that after state / (say) it goes 

 to statey (say) and not to any other of its states. This is the ordinary 

 selection that a machine makes when its "message" (the protocol 

 from it) says that the machine has this transformation and no other. 



Another process of selection shown by a machine is that noticed 

 in S.7/24: every determinate machine shows selection as it reduces 

 the variety in its possible states from the maximum initially to the 

 number of its basins finally. 



Another process of selection was treated in S.5/13, when one part 

 of a whole can select from states of equilibrium in the other part 

 by "vetoing" some of them. Tliis is perhaps the most obvious form 

 of selection, for, as the two are watched, the imaginative observer 

 can almost hear the vetoing part say ". . . no good, still no good, I 

 won't have it, still no good, Hold It! — yes, we'll keep that per- 

 manently." If a machine is to be built as a selector (perhaps to carry 

 out the programme hinted at in the final section) it will, so far as I 

 can see, have to be built to act in this way. It is the way of the 

 second-order feedback in Fig. 5/14/1 (supplemented in S.12/15). 



There are doubtless other methods, but these will suffice for illus- 

 tration, and they are sufficient to give definiteness to the idea of a 

 machine "selecting"; (though special consideration is hardly 

 necessary, for in Shannon's theory every act of communication is 

 also one of selection — that by which the particular message is 

 caused to appear). 



13/20. Duration of selection. At this point a word should be said 

 about how long a given act of selection may take, for when actual 

 cases are examined, the time taken may, at first estimate, seem too 

 long for any practical achievement. The question becomes specially 

 important when the regulator is to be developed for regulation of a 

 very large system. Approximate calculation of the amount of 

 selection likely to be necessary may suggest that it will take a time 

 far surpassing the cosmological; and one may jump to the conclusion 

 that the time taken in actually achieving the selection would have to 

 be equally long. This is far from being the case, however. 



The basic principles have been made clear by Shannon, especially 

 in his Communication theory of secrecy systems. He has shown that 

 if a particular selection is wanted, of 1 from A'", and if the selector 

 can indicate (or otherwise act appropriately) only as to whether the 

 required element is or is not in a given set, then the method that 



260 



