REGULATING THE VERY LARGE SYSTEM 13/8 



in that minute. There is also the long interval. After it has worked 

 for a year someone may ask me whether it has proved a good regula- 

 tor over the year. While deciding the reply, I think of the whole 

 year's disturbance as a sort of Grand Disturbance (made up of many 

 individual disturbances, with a small d), to which it has produced a 

 Grand Response (made up of many individual responses, with a 

 small r). According to some standard of what a bath should do 

 over a year (e.g. never fail badly once, or have an average deviation 

 of less than -J", etc.) I form an opinion about the Grand Outcome — 

 whether it was Good or Bad — and answer the question accordingly. 



It should be noticed that what is "Good" in the Grand Outcome 

 does not follow necessarily from what is "good" (tj) in the individual 

 outcomes; it must be defined anew. Thus, if I go in for a lottery 

 and have three tickets, a win on one (and consequent loss on the 

 other two) naturally counts as "Good" in the Grand Outcome; so 

 here I good + 2 bad = Good. On the other hand, if I am tried 

 three times for murder and am found not guilty for one, the indi- 

 vidual results are still I good + 2 bad, but in this case the Grand 

 Outcome must naturally count as Bad. In the case when the indi- 

 vidual disturbances each threaten the organism with death. Good 

 in the Grand Outcome must naturally correspond to "good in 

 every one of the individual outcomes". 



These Grand Disturbances are vectors whose components are the 

 individual disturbances that came hour by hour. These vectors 

 show a form of constraint. Thus, go back to the very first example 

 of a vector (S.3/5). It was A ; contrast it with B: 



A B 



Age of car: Age of Jack's car: 



Horsepower: ,, ,, Jill's ,, 



Colour: „ „ Tom's ,, 



Obviously B is restricted in a way that A is not. For the variety 

 in the left-hand words in ^'s three rows is three; in 5's three rows it 

 is one. 



Vectors like B are common in the theory of probability, where 

 they occur under the heading "sampling with replacement". Thus, 

 the spin of a coin can give only two results, H or T. A coin spun 

 six times in succession, however, can give results such as (//, //, T, 

 H, T, H), or (T, T, H, H, T, H), and so on for 64 possibilities. 

 (Compare S.9/9.) 



What is important here is that, in such a set of vectors (in those 

 whose components all come from the same basic class, as in B), 



249 



