388 John R. Platt 



In the adult, the need for continuous redetermination of addresses becomes 

 smaller. As Dr. W. A. Arnold of the Oak Ridge National Laboratory pointed 

 out when this paper was first given, the words on a printed page that is illuminated 

 for only 10""* sec, too short for any scanning, can be read (by anyone who 

 can read) quite normally over an area comparable with the foveal diameter. 

 The addresses have already been established and need little if any reconfirmation 

 from the operations that generated them. 



How complete and rigid they are might be discovered if we knew the limits 

 of developmental distortion of the retina after, say, the age of 3. Or if 

 adults could work with optical systems giving subtle distortions of a few 

 minutes of arc in the shape and topology of patterns in the foveal region, 

 to see the effect of the loss of addresses on line, circle and pattern perception, 

 on reading and the identification of persons, and how soon — and how far 

 back in the network — a new set of addresses would be learned. 



IV. NECESSARY PROPERTIES OF NON-ADDRESSED SYSTEMS 



Certain properties would seem to be necessary characteristics of at least 

 the early stages of all non-addressed and address-determining networks. 

 What is interesting is that many of these seem to be familiar aspects of higher 

 human behavior, restated in receptor-network terms. We may properly inquire 

 how far our more complex activities can be subsumed under a generalized 

 functional geometry, and how far our more complex experiences are organized 

 by means of generalized displacement congruences. 



A. Operational Characteristics 



The null-transmitter delay cell which it seems necessary to invoke for any 

 network making time-delay comparisons of patterns would be a suitable 

 prototype for much of our higher neural organization. In the mature organism, 

 after the structure and connections of such a cell have been stabilized — that 

 is, after the addresses of its input channels have been determined — the cell 

 will always collate two or more input patterns in a standard way to produce 

 a simplified vital output. Let us focus attention, first, on the nature of the 

 outputs, then on the inputs, and finally on the process as a whole. 



Abstraction of Invariant Pattern Properties 



The output of each such cell signals to the higher stages the presence of 

 some particular kind of simple or complex pattern in the lower stages. This 

 implies in turn a pattern in the first-stage images, representing a pattern in 

 the external field. The process is abstraction. Pattern is another name for 

 congruences or invar iances in this field. 



At a given instant, some fifth or tenth stage cell may be signaling That 

 is the letter R'. Simultaneously some other set of elements and delay lines 

 is abstracting from the same retinal elements the information 'It is in my 

 wife's handwriting'. A third neuron connected to these elements says, 'It is 

 black'; a fourth, 'It is large'; and so on, for details and context and background 

 and all the other components. Perhaps some still higher neuron also signals 

 the unification of these separate pattern properties and others into a word. 



