374 John R. Platt 



manufacture. Sooner or later the increase with increasing m will limit the size 

 of the pre-addressed systems we can construct, no matter how much the assembly 

 process is speeded up. 



But for a non-addressed system, even with 10^ or 10^ elements, a very few 

 specifications of the general assembly or growth patterns may suffice (3). The 

 construction is cheaper, whether measured in assembly information, in time or 

 money. Obviously there is a price. It is that the addresses of all retinal elements 

 must now be learned— after operations begin. The construction is speeded up; 

 the attainment of full operating efficiency is delayed until address-determination 

 is completed. But the non-addressed system constructed with a given amount 

 of assembly information can eventually become far more complex and 'intelligent' 

 than its pre-addressed counterpart. 



This initial incompetence may be why, in evolution, the non-addressed 

 organisms only become prominent when parental care appears in family systems 

 like those of birds and mammals. The long learning time for large m might be 

 connected with the long childhood of the more intelligent species. 



Actually there may be no sharp boundary in biology between the pre- 

 addressed and the non-addressed. On evolutionary grounds alone, a vitally 

 necessary fraction of the human brain must certainly be pre-addressed. The 

 autonomic nervous system may be largely so constructed. Reflex actions and 

 probably color vision seem to have this character. The non-addressed sections 

 of our networks, although perhaps responsible for our most characteristically 

 human activities, may be a late and still secondary addition to a large pre- 

 addressed core — as Dr. Sacher stressed in his comments on this paper. 



It is often asserted that nerves and synaptic connections do not grow. This 

 might be true for the pre-addressed sections; but it should be false for the 

 non-addressed sections. Address-learning in a network necessarily means 

 creation or change of connections. Change of neural connections means growth 

 or atrophy or both. If new synaptic connections do not grow, they must at 

 least be selectively and permanently activated or deactivated during the address- 

 detennining process. 



The Pattern Question 



Whatever the economy of assembly, the question remains: Can randomly 

 arranged elements be used to make discriminations of straight lines or of any 

 other types of pattern elements ? 



There is evidently an intimate connection between the perception of pattern 

 and the determination of the addresses of the retinal elements. To make the 

 question more precise, let us number the elements 123 • • -y • • • in as nearly the 

 same way as possible in all retinas, and set up coordinate axes as nearly alike 

 as possible. The randomness means that element y will have different address 

 coordinates, X;, j^ in every retina. Or better, we might specify addresses by 

 relationships rather than coordinates, giving them forms such as 'Element y is 

 collinear between elements g and p\ This address might be right in one retina, 

 wrong in another. 



Such an uncertainty of internal pattern has to be resolved within the network. 

 The question is then: Can the coordinates x^y^ be detennined, or can the 

 straight-line or other geometrical spatial relations of element j to many other 



