SCIENTIFIC DISCOVERY 181 



way that a block of marble is the potentiality of the result- 

 ing statue. 



LOGIC OF DISCOVERY 



The precise sense in which there can be said to be a logic 

 of discovery should be made clear at the outset. It seems 

 safe to say that there is no deductive logic of discovery ; data 

 do not contain their own explanatory hypotheses in the 

 same way that premises contain their conclusions. Even in 

 deductive logic there exists the important problem of how 

 the conclusion of a deductive argument can be both novel and 

 true. To the extent to which it seems to be new, it cannot be 

 contained in the premises; hence it cannot be implied by 

 them. But to the extent to which it seems to be true, it must 

 be contained in the premises; hence it cannot exhibit novelty. 

 The usual way of avoiding this difficulty is to deny that 

 deductive conclusions have any genuine novelty. But 

 hypotheses obtained by inductive discovery are obviously 

 new, and one's only recourse, therefore, is to deny that they 

 have necessary relations with the data. It is therefore 

 impossible to look upon inductive discovery as a process 

 analogous to the operation of a computing machine; one 

 should hardly expect to pour data into one spout, turn a 

 crank, and grind out an explanatory hypothesis from another. 



Yet there is clearly some sense in which data may be said 

 to "contain" their own explanatory hypothesis. Hypotheses 

 are not drawn from the blue by purely creative acts of mind. 

 On the contrary they are determined by data in at least 

 three ways. They are determined in the sense that if there 

 were no data there would be no hypothesis and if the data 

 were different the hypothesis would be different. They are 

 determined by a more inclusive range of facts in the sense 

 that every hypothesis is probably based to some extent on 

 an analogy. They are determined, finally, by the general 

 nature of the explanatory relation, i.e., by the fact that the 

 symbols which describe the hypothesis must be such as to 

 imply the symbols which describe the data. 



