Plate 1. First, let one eye be closed, and let the observer rotate his open eyeball to and fro by pressing as lightly as possible on the outer corner of the eyelid, while looking at the figure. A curious non-Euclidean distortion of the pattern will be seen ( MacKay, 1958). The rays lying nearest to the direction of displacement of the image appear to diverge and converge as if the figure were on an elastic surface being locally distended and contracted. If, however, a thread or other marker is laid across the direction of image-displacement to one side of the centre, the distortion in its neighbourhood is suppressed, although on the opposite side (without a marker) it persists unaltered. † The details of the illusion do not concern us now, and the perceptual process involved is unlikely to lie in the primary visual cortex; it is intended only to illustrate the apparent need for a continuous model, incorporating such features as elasticity.
The final illustration also depends on Plate 1. If the pattern is inspected for about ten sees. in a good light, and then quickly replaced by a blank surface, the latter appears to be traversed briefly by a circular swirl of shadowy, wavy lines whose general direction is roughly perpendicular to the lines of the stimulus figure. By superimposing on Plate 1 a background of random visual 'noise' (a succession of dots in randomly varying positions) it has proved possible to excite perception of what I have called the 'complementary image' continuously--rather in the way that iron-filings reveal the state of strain around a magnet ( MacKay, 1957). Any pattern with more than, say, four to eight roughly parallel lines has been found to generate a neural state of strain, forming a moving complementary image of this sort.
Once again we need not stay now with details. It would seem however that this kind of persistent and powerful long-range interaction, occurring between widely separated areas of the visual field, and independent of position on the retina, bespeaks the type of continuous model we have been discussing, incorporating essentially mobile and wave-like patterns of excitation.
Can any conclusions be drawn at this stage in a rapidly-moving process? Digital computing mechanisms seem to have had their day as models of complex neural systems, though digital computors used as simulators of more realistic models may well be indispensable as complexity increases. Analogue simulators too--as Taylor ( 1956) and others have shown--are likely to be of increasing value, while circuits that grow and function under statistical feedback should fill an important gap in our model-making repertoire. The only additional suggestion in this paper is that in all such models we seem to be already too near the limits of practicability if we insist on using discrete circuit-elements and discrete connexions. For systems of multi-million (or____________________