6/5 AN INTRODUCTION TO CYBERNETICS 



Ex. 6/3/1, that ay is always followed by either a/ or /S/^that although 

 the a's transition is not single-valued, that of they is. 



So he examines the record. Usually his first concern is to see 

 whether the Box is absolute if the input state is given. He does 

 this by collecting: 



(i) all the transitions that followed the input state a, sorting 

 them into what g went to, what // went to, and so on through 

 all the output states; 



(ii) the same for input j8; 



(iii) and so on through all the observed input states. 



What he tries, in other words, is to fill in a set of transformations 

 like those of S.4/1, and he examines what he gets to see if they are 

 single-valued. 



Thus, if the given protocol is tested, and if every one of the 16 

 transforms is recorded, there results: 



(No transition was observed from g with input at j8.) Within each 

 cell the letters are all equal, so the table can be simplified to: 



with a statement that throughout the protocol this closed single- 

 valued transformation was observed. 



Thus by direct re-coding of the protocol the experimenter can 

 demonstrate that the behaviour is machine-like, and he can deduce 

 its canonical representation. 



It should be noticed that he has deduced it from direct observation 

 of the Box's actual behaviour. He has relied on no "borrowed" 

 knowledge. Whatever he may have expected, and regardless of the 

 confidence of his expectation, the final deduction depends only on 

 what actually happened. Thus, in any conflict between what he, or 

 others, expected and what was found, these empirical results are 

 final as a statement of the Box's nature. 



Should the system not be determinate, i.e. the transformation not 

 single-valued, he can proceed in either of two ways. 



90 



