232 A. Rapoport 



theory into psychology in a significant way, not as a mere appendage to statistics. 

 I think the situation is similar in the case of information theory. Instead of 

 lamenting the ambiguity of the 'universe of discourse' in psychological situations, 

 which stands in the way of a straight-forward appHcation of infonnation theory, 

 we may well seek to infer the universe of discourse as it looks from inside the 

 subject. This is, indeed, a central task of the psychologist, and it is improper 

 for him to shun it. 



It remains true, however (and it is just as it should have been), that the early 

 psychological experiments based on information-theoretical considerations 

 were constructed in such a way as to eliminate idiosyncratic subjective probabili- 

 ties. In memory tasks one starts with some set of thoroughly randomized and 

 presumably equalized stimuli. One introduces redundancies in terms of actual 

 biases of occurrence-frequencies objectively determined. The same techniques 

 prevail in experiments in which the capacity of the individual as a channel is 

 measured. These are all attempts to translate into experimental psychology 

 situations occurring in communication engineering. The human being is studied 

 as a piece of communication apparatus. I beheve this strategy to be entirely 

 correct as far as it has gone. I am sure, however, that its limitations are apparent 

 most of all to the investigators who pursue it. Somewhere along the line a 

 transition must be made which will allow the application of information theory 

 to psychology as distinct from psychophysics. In other words, the perceptual 

 world of the subject must eventually become a focus of interest. There is no 

 reason why information theory should not become a useful tool in such investiga- 

 tions. 



Characteristically, the tasks just described (such as rote learning, multiple 

 choice responses, and so on) do not involve the deductive process. Indeed, 

 from the formal information-theoretical point of view, the results of infomiation 

 theory are not applicable to a deductive process, because there is no 'uncertainty' 

 in such processes. The solution of a mathematical problem, no matter how 

 complex, yields no information from the information-theoretical point of view. 

 From either the common-sense or the psychological point of view such a con- 

 clusion seems bizarre. The information-theorist can, of course, argue that his 

 technical definition of information departs from common-sense and psycho- 

 logical notions of what constitutes infonnation, and he is technically correct. 

 Yet it might be instructive to try to bring the two concepts of 'information' 

 into closer agreement. 



Formally speaking, no information is gained in the solution of a purely 

 mathematical or logical problem, because the solution is implicitly contained in 

 the already known conditions. But the solution is not initially known to the 

 subject. Is there a way to measure the extent of his ignorance ? There might 

 be, if we are willing to abandon the omniscient position from which the solution 

 is seen as a necessary inference, hence of zero uncertainty, and enter the per- 

 ceptive of the cognitive field of the subject, to whom only a range of possible 

 solutions and, perhaps, associated subjective probabilities present themselves. 

 But how does one get into this perceptive field? Obviously by observing the 

 subject's behavior. But how does one make inferences from the observations to 

 what that perceptive field may be ? It would be gratifying to be able to say that 

 for every configuration of the cognitive field, there is a specific behavior pattern, 



