390 John R. Platt 



impossible process of applying m independent abstracted relations simul- 

 taneously could the m independent inputs be inferred. But this consumption 

 or 'loss' of information is not biological loss but a gain, since it represents 

 the selection of the biologically relevant item from the confusing irrelevant flux. 

 After a lifetime of suppression of the less valuable pattern and field details, 

 adults finally attend only to the later neuron outputs or abstractions and seem 

 to lose the eidetic ability to bring forth the exact early-stage patterns of instan- 

 taneous retinal excitation, except as they can reconstruct them approximately 

 from their appropriate or inappropriate collection of output neurons. This 

 may show the cessation of early-stage rearrangements, which finally become 

 completely pre-addressed as far as new experiences are concerned. 



Analogy Perception 



We may look not only at what has been abstracted — the outputs — but 

 at what has been compared — the inputs. The elementary process in address- 

 determination was the comparison of excitation patterns at two different times. 

 A network whose neurons can signal identities or similarities of pattern — dis- 

 placement congruences — is an analogy-perceiving network. Much, if not all, 

 of what we call intelligence may be the abihty to perceive successive analogies 

 at higher and higher levels of abstraction, a multiple repetition of a single basic 

 neural process of organization. 



Artificial pre-addressed systems are not generally able to perceive any 

 analogies except between those sets of inputs that they are wired up to treat 

 as equivalent. The value of mechanical mosaic detectors such as the punch- 

 card reading-head lies in the fact that they are wired up to perceive obscure 

 informational analogies and not any of the space or time pattern analogies of the 

 kind that we perform easily in retinal abstraction. 



Thoughts and Symbols 



Pattern and analogy perception resemble some important aspects of the 

 higher-order process we call thought. We might make a limited definition of 

 a thought as the realization of previously unperceived pattern-relationships. 

 A thought could then be represented by an operator equation, 



where P^" is a pattern of the «th stage of abstraction ; Pg"* is one of the mih. 

 stage; q is the time delay or other transformation operator which relates 

 pattern P^ to P^. And Qph the {n + l)st or {m + l)st (whichever is higher) 

 stage of relationship; it is the pattern of the P's, a pattern of patterns. It is 

 primarily a realization signal — a displacement-congruence signal — but it 

 may also contain some or all of the common elements of the P patterns. We 

 might distinguish if necessary between the possibility of the thought and the 

 continued existence of the thought-relationship; and between the insight, 

 or first assertion of the thought, and the repeated use of the established thought. 

 If P2'"* has no perceivable relationship to P-^", there is no thought, 



Po^qP^' = for all q. 



