Functional Geometry and the Determination of Pattern in Mosaic Receptors 397 



balance among second-stage detectors of different kinds, and of foveal versus 

 peripheral vision, the role and mechanism of fixation and attention, and the 

 whole output-selection problem which has been ignored here. Theoretical 

 consideration might lead to principles of neural connection, unused in biological 

 systems, which would produce entirely different kinds of 'intelligence' in the 

 organization of the input fields. 



Actual construction of at least the first stages of a non-addressed system 

 might even be relatively easy. Since the receptor elements do not need to be 

 wired individually, they can be laid down en masse, like the 10^ crystals of a 

 photographic emulsion. The first-stage neuron layer, second-stage layer, and 

 so on, could be laid down similarly. The crystals could not be compact in 

 shape, like those of the emulsion, but would have to be interbranching needles. 

 But the first successful device might be many orders of magnitude more complex 

 than anything now made. 



To create such a device would require a number of really penetrating 

 chemical or electrical inventions, but perhaps not a prohibitive number. 

 Oculomotor outputs for scanning and tracking might have solutions close 

 to the present standard single-element solutions. The main problem would be 

 to guarantee that the neuron connections will tend to grow in such directions 

 as to support any congruences in the chemical or electrical time-patterns, 

 and will tend to be dissolved otherwise. 



With elements having 10~^ second time-constants (comparable to transistors) 

 the potential learning speed of such a device would be 10^ times faster than that 

 of a human being (2 hours = 20 years). Such speeds could not be fully realized 

 because the initial address-determination will be limited by scanning speeds 

 and motor-output speeds and by the chemical speeds of deposition of successive 

 layers. But these potential speeds and these limitations are comparable to 

 those of a digital computer; the latter being similarly held back by the slowness 

 of programming and input and by the slow storage access speeds. 



A pattern-perceiving device so much faster than a human being and with 

 a full range of inputs and outputs would pose grave problems of education, 

 manipulation and control, problems different from those of a digital computer 

 and more difficult; but the rewards would be correspondingly greater if these 

 problems could be solved. 



Acknowledgement — This study was stimulated by a number of conversations 

 with Professor Richard L. Meier, now of the University of Michigan. 



REFERENCES 



1. J. R. Platt: Amplification aspects of biological response and mental activity. Amer. 

 Scient. 44, 180-197(1956). 



2. S. L. Polyak: The Retina, Fig. 40, University of Chicago Press, Chicago, III. (1941). 



3. W. C. Pitts: Proc. Western Joint Computer Conference at Los Angeles, 1-3 March 1955, 

 108-110, Inst. Radio Engng., New York (1955). 



4. J. Strong: Procedures in Experimental Physics, Prentice-Hall, New York (1938). 



5. H. A. Rowland: The Screw. Encyclopedia Britannica, 1 1th ed. (1913). 



6. J. Strong: New Johns Hopkins ruling engine. J. Opt. Soc. Amer. 4i, 3-15 (195\). 



