trees showed no essential differences: the more branches, the greater the dynamic range. Psychophysical experiments by von Bekésy (ref. 7) on skin vibration sensitivity confirm these results. In skin areas with dense innervation (many branches per unit area) greater dynamic range and slower rising subjective magnitude ("loudness") curves result than in sparsely innervated areas. Conversely, in a given area, a small diameter vibrator (exciting only part of a neuron tree) gives rise to steeper loudness functions and less range than a larger vibrator. CONCLUSION These studies with neuromimes have, therefore, yielded some interesting insights into various properties of the peripheral nervous system. On the one hand, by implementing the known operations of the frog's retina, we have gotten some idea about the importance of certain anatomical features of the retinal ganglion cells. On the other hand, by imitating an anatomical arrangement, we found a novel function, and by extension have found an explanation of some as yet insufficiently explained phenomena such as recruitment and "loudness curves" in the skin. In answer to the question which is the title of this paper, we feel we can state that we have demonstrated how neuromimes can be used rather effectively to study nervous processes "by proxy." How far one can go along this line remains an open question which can only be answered by further work. The initial successes are certainly heartening, but we are convinced that, unless we hew closely to the original living system, and continually verify our deductions, we may follow false leads and wind up ins Blaue, without footing in either biology or engineering. REFERENCES 1. van Bergeijk, W. A., Nomenclature of Devices Which Simulate Biological Functions. Science, 1960 (in Press) 2. Harmon, L. D., Artificial Neuron. Science, 129: 962-963 (1959). 3. Lettvin, J. Y., Maturana, H. R., McCulloch, W. S., and Pitts, W. H., "What the Frog's Eye Tells the Frog's Brain." Proceedings IRE, 47: 1940-1951 (1959). 4. Fernandez, C., The Innervation of the Cochlea (Guinea Pig) Laryngoscope, 61: 1152-1172 (1951). 5. Lorente de No, R., The Sensory Endings in the Cochlea, Laryngoscope, 47: 373-377 (1937). 6. Bogert, B. P., Determination of the Effects of Dissipation in the Cochlear Partition by Means of a Network Representing the Basilar Membrane. J. Acoust. Soc. Am., 23: 151-159 (1951) 7. Beke sy, G. von, Funneling in the Nervous System and its Role in Loudness and Sensation Intensity on the Skin. J. Acoust. Soc. Am., 30: 399-412 (1958); Neural Funneling Along the Skin and Between the Inner and Outer Hair Cells of the Cochlea. J. Acoust. Soc. Am.. 31: 1236-1249 (1959). WADD TR 60-600 TECHNICAL SESSION IV MECHANICAL REALIZATION OF Moderator: Dr. David O. Ellis Litton Industries INFORMATION PROCESSING LANGUAGES AND HEURISTIC W.R. Reitman Carnegie Institute of Technology INTRODUCTION "Das Glasperlenspiel" is a novel by Hermann Hesse, published in 1943 and later awarded a Nobel Prize. In it, Hesse imagines the development by scholars and scientists of a generalized symbolism--the "Bead Game", "capable of expressing the contents and results of nearly all the sciences and of placing them in relation to each other" (1). Recent work on problem oriented common languages (Bauer and Samelson, 2) suggests that digitial computers, as "general information transformers, or symbol manipulators" (Gorn, 3) may one day provide a basis for such symbolisms. Hesse's ideas are of course rather remotely related to current work on common languages for numerical analysis and scientific computation. Similarly, those using the techniques discussed here in studies of intelligence in natural and artificial systems can claim only a collateral relation to the Bead Game players. For in the first place, Hesse wanted a universal language in which to encode the knowledge man achieves; we on the other hand are at least as interested in discovering how he goes about achieving it. Secondly, Hesse sought universality in a set of basic formulas, relations, and analogies presumed to underlie many different realms, whereas we expect similarities in problem solving to derive from a common source of constraints: the information manipulating capacities of the human mind. Still, there is a certain family resemblance in aspirations, and perhaps this will serve as sufficient excuse for our title. LIST PROCESSING AND HEURISTIC PROGRAMMING Developed to provide flexibility and power in manipulating nonnumeric symbols, list processing languages provide easy access by name to information organized in complex list structures of any size. The language conventions are designed so that such information structures may be searched, compared, or reorganized without advance knowledge of their arrangement or physical location. Programs may be recursive, and may involve arbitrarily complex hierarchies of subroutines. The cost in space and time is high; but having paid the price, we are able to deal with systems which otherwise would be quite inaccessible. Several of these languages are described in some detail in recent articles (4,5, 6,7), and a general discussion of symbol manipulation (8), by Green, will be available shortly. aThis paper draws extensively on the work of A. Newell, J.C. Shaw, and H. A. Simon. Invaluable discussions with Dr. Newell during the writing of the paper also are gratefully acknowledged. Finally, the writer wishes to thank the Social Science Research Council for a Faculty Research Fellowship which made possible his work in this area. We can at this point give no good definition of heuristic programs as a class. They are, after all, first attempts to incorporate in computer programs processes analogous to those used by humans in dealing intelligently with ill-structured problems; and we are only beginning to understand what these processes are. In this way, heuristic programs are distinct from algorithms, which guarantee solutions to problems just exactly because the designers have anticipated every permissible path, and have incorporated machinery which can cope with each of them. A heuristic system also may possess a great deal of knowledge of its problem area; but it must possess routines enabling it to deal adaptively with the unexpected. In addition to their uses in representing and studying complex information processing, heuristic programs also find application as tools for dealing with problems for which no algorithms exist. Their use in such cases may in fact facilitate the design of suitable algorithms by clarifying the structure of the problem area. APPLICATIONS Investigations employing information processing languages and heuristic programming include studies of problem solving in mathematics and symbolic logic, in industrial and business problems, in laboratory tasks, in chess playing, and in the composition of music. These techniques also serve as a basis for investigations of the understanding of language and for attempts to develop programs which deal intelligently with many different kinds of problems, as humans do. We survey these applications briefly, limiting ourselves primarily to problem solving, since learning and pattern recognition programs are discussed elsewhere in this session. We then consider in somewhat more detail current and projected work on general problem solving systems (GPS) and on a program which composes music. Using these two projects as vehicles, the paper concludes with a discussion of some longer range problems, strategies, and aspirations for work with these techniques. The earliest heuristic program was the Logic Theorist (Newell, Shaw and Simon, 9), which made use of methods and rules of thumb derived from observations of humans solving problems in the sentential calculus. Newell, Shaw and Simon lay special stress on the heuristic importance of the matching and similarity routines realized in the program. LT succeeded in proving 38 of the 52 theorems it was given, and in one case actually found a proof more elegant than Whitehead and Russell's (10). Gelernter's program for proving theorems in geometry (11) is another major achievement in this area. It makes use both of a "syntax computer", related to the syntactic heuristics of LT, and of a "diagram computer", which contains "a coordinate representation of the theorem to be established together with a series of routines that produce a qualitative description of the diagram." The program rejects statements which do not hold in its diagrams, and so the system behaves much as if it could draw figures, scan them, and discriminate intelligently on the results. Related to these efforts is a program under |