148 ANNUAL REPOET SMITHSONIAN" INSTITUTION, 19 3 2 



was silver nitrate. It is no longer silver nitrate after our treatment 

 of it. This is an example of retrospective inference : The property 

 which we infer is not that of " being X " but of " having been X." 

 We noted at the outset that in considering determinism the alleged 

 causes must be challenged to produce their birth certificates so that 

 we may know whether they reall}^ were preexisting. Retrospective 

 inference is particularly dangerous in this connection because it 

 involves antedating a certificate. The experiment above mentioned 

 certifies the chemical constitution of a substance, but the date we 

 write on the certificate is earlier than the date of the experiment. 

 The antedating is often quite legitimate; but that makes the prac- 

 tice all the more dangerous, it lulls us into a feeling of security. 



RETROSPECTIVE CHARACTERS 



To show how retrospective inference might be abused, suppose 

 that there were no way of learning the chemical constitution of a 

 substance without destroying it. By hypothesis a chemist would 

 never know until after his experiment what substance he had been 

 handling, so that the result of every experiment he performed would 

 be entirely unforeseen. Must he then admit that the laws of 

 chemistry are chaotic? A man of resource would override such a 

 trifling obstacle. If he were discreet enough never to say beforehand 

 what his experiment was going to demonstrate, he might give edify- 

 ing lectures on the uniformity of nature. He puts a lighted match 

 in a cylinder of gas and the gas burns. " There you see that hydro- 

 gen is inflammable." Or the match goes out. " That proves that 

 nitrogen does not support combustion." Or it burns more brightly. 

 " Evidently oxygen feeds combustion." " How do you know it 

 was oxygen?" "By retrospective inference from the fact that the 

 match burned more brightly." And so the experimenter passes from 

 cylinder to cylinder; the match sometimes behaves one way and 

 sometimes another, thereby beautifully demonstrating the uniform- 

 ity of nature and the determinism of chemical law. It would be 

 unkind to ask how the match must behave in order to indicate 

 indeterminism. 



If by retrospective inference we infer characters at an earlier date 

 and then say that those characters invariably produce at a future 

 date the manifestation from which we inferred them, we are working 

 in a circle. The connection is not causation but definition, and we 

 are not prophets but tautologists. We must not mix up the genuine 

 achievements of scientific prediction with this kind of charlatanry, 

 nor the observed uniformities of nature with those so easily invented 

 by our imaginary lecturer. It is easily seen that to avoid vicious 



