Machines and Societies 
predict them; and in an age that takes verification as its chief 
criterion of truth the manipulators could have the strongest 
possible defence: “‘We were right, weren’t we?” 
THE LIMITS TO MACHINES AS MODELS 
Behind this curious situation there is in fact a basic logical 
impasse. Any complete description of a cognitive information- 
system must include, or depend on, the information possessed 
by the units of the system. Any change in the information 
possessed by a unit must, in general, require a change in the 
complete description. It follows that in general, no complete 
description exists which would be equally valid whether or not the units 
were informed of it. In other words, no complete machine-model 
(nor any other complete predictive model) of a society is 
possible, which would be equally valid before and after any 
member of that society learned of it. In this area, then, there 
is a fundamental incompatibility between two of the normal 
aims of science--to observe facts, and to spread knowledge of 
those facts as widely as possible. 
The implications of this could take us far afield. Here we 
must simply note a few consequences. In the first place, the 
situation of society is not quite as desperate as it might seem, 
for any would-be predictor of society who is sufficiently in- 
volved in that society must find his calculations frustrated by 
ignorance of his own (future) reactions to his own (as yet 
incomplete) conclusions. No improvements in his computing 
facilities can obviate this fundamental limitation. Moreover, 
even if he were to isolate himself for a time, his potential victims 
could indefinitely embarrass him by equipping themselves with 
a similar predictive model. 
Secondly, on many questions of social attitude now open to 
scientific study, it is fallacious to suppose that there must exist 
neutral scientific knowledge to be publicly acquired. The 
declared aim of science is to propound conclusions which are 
true regardless of the attitude people take to them. It is now 
abundantly clear that many questions being asked of applied 
social science even today have no such answers. To recognize the 
165 
