12/13 AN INTRODUCTION TO CYBERNETICS 



Another example of regulation, of a low order of efficiency, would 

 be shown by a rat with serious brain damage who cannot remember 

 anything of a maze, but who can recognise food when encountered 

 and who then stops to eat. (Contrast his behaviour with that of a 

 rat who does not stop at the food.) His progression would be 

 largely at random, probably with some errors repeated; nevertheless 

 his behaviour shows a rudimentary form of regulation, for having 

 found the food he will stop to eat it, and will live, while the other rat 

 will keep moving and starve, 



Ex. ] : A married couple decide to have children till they have a boy and then 

 to stop, (i) Is the process regulatory? (ii) What is the matrix of transition 

 probabilities ? 



Ex. 2: Is the game "Heads, I win; Tails, we toss again" regulatory? 



12/13. So far we have considered only the way in which a 

 Markovian machine moves to its goal. In principle, its sole 

 difference from a determinate machine is that its trajectory is not 

 unique. Provided we bear this difference in mind, regulation by the 

 Markovian machine can have applied to it all the concepts we have 

 developed in the earlier chapters of this Part. 



(The warning given in S. 11/11 (para. 5) must be borne in mind. 

 The steps that take a Markovian machine along its trajectory are of 

 a smaller order of magnitude than the steps that separate one act 

 of regulation (one "move" in the sense of S.11/3) from another. 

 The latter steps correspond to change from one trajectory to another 

 — quite different to the change from one point to the next along one 

 trajectory.) 



Thus the basic formulation of S.11/4 is compatible with either 

 determinate or Markovian machines in T and R to provide the 

 actual outcome. No difference in principle exists, though if we 

 describe their behaviour in psychological or anthropomorphic terms 

 the descriptions may seem very different. Thus if R is required (for 

 given disturbance) to show its regulatory power by going to some 

 state, then a determinate R will go to it directly, as if it knows what 

 it wants, while a Markovian R will appear to search for it. 



The Markovian machine can be used, hke the determinate, as a 

 means to control; for the arguments of S. 11/14 apply to both (they 

 were concerned only with which outcomes were obtained, not with 

 how they were obtained.) So used, it has the disadvantage of being 

 uncertain in its trajectory, but it has the advantage of being easily 

 designed. 



232 



