Remarks on the Problem of the Logical Design of the Vertebrate Visual System

Most interest ing machine tasks, e.g., v isual pattern recognit ion, requi re large complex systems. Conventional machine designs are not we l l suited to the re l iab i l i t y demands of large complex systems. Many designs, conceived and tested on a smal l scale, and theoret ical ly extendable to large scale systems, are s t i l l awkward and imprac t ica l as a large scale system because re l iab i l i t y appears to vary inversely wi th the number of components. Animal systems seem to have solved this problem and offer some hope for understanding the problem of how to bui ld large complex systems with high re l iab l i ty . In par t icu lar animal systems seem to to lerate a great deal of component var iat ion and noise. It is suggested that animal systems represent designs that actually take advantage of noise, unlike conventional machines, and that some of these animal systems may provide techniques which are applicable to the problem of construct ing re l iab le machines f rom unrel iable components.