201
THE KOLE OF MODELS IN THEORETICAL BIOLOGY
human perception and has explored heuristic Computer programs to simulate this process.
Advanced mathematical control theory has been applied to adjust-ments of the pupil and to positioning of the eyeball. Stark (1959) showed that servomechanism invariants could be used to analyze and describe the natural control mechanisms of the pupil when presented with fluc-tuating light levels. In this instance a numerical or electroanalog partial model of the pupil was sufficiently precise to predict previously un-recorded responses in some instances. The appropriate invariants for this case include “gain,” “feedback factor,” etc.
Servomechanism and discrete control mechanism theory have also been used with success for analysis of movements of the human eyeball when tracking an object in the visual field (Young and Stark, 1963; Johnson and Fleming, 1963; Wammeunde, 1964). Over-all performance scores would appear to be the pertinent similarity invariants for the complex Computer models of eye tracking. Similar methods have been used to create models of human manual-tracking action (Licklider, 1960; Adolph, 1959; Houk and Stark, 1962), such as is reąuired for guidance of airplanes or space vehicles. Morę ambition simulations use Monte Carlo methods (Adams and Webber, 1963) or advanced control theory (Bekey, 1963) to produce adaptive or leaming behavior.
It thus becomes elear that considerable success has been achieved in qualitative and semiquantitative simulation of neurophysiological regulatory mechanisms by use of existent mathematical control theory methodology. In all cases, the models are restricted in naturę and have only rarely predicted new facts, but they and their natural prototypes can be compared using one or morę common invariants of the performance score variety.
Models of this class are characterized by similarity invariants per-taining to algorithmic structure rather than physical or numerical design, i.e., they function by executing the same set of rules as the natural system, but are not similar in regard to physical mechanisms for im-plementing these rules. If an algorithm is defined as a “definite computa-tional procedurę,” then an algorithmic model is one which uses the same algorithm as the prototype. In some instances there is not even this degree of similarity, and the model and prototype simply get the same result in a test situation, being similar in regard to a performance crite-rion of some circumscribed variety.