|
| Selection by Consequences: A Computational Theory of Behavior Dynamics |
| Monday, May 26, 2008 |
| 3:30 PM–4:50 PM |
| Barbershop |
| Area: EAB/TPC; Domain: Theory |
| Chair: Jack J. McDowell (Emory University) |
| Abstract: A computational theory of selection by consequences that implements Darwinian principles of selection, reproduction, and mutation with respect to a population of potential behaviors yields equilibrium behavior in single and concurrent schedules that is consistent with matching theory. In addition to its relevance for behavior analysis, this theory may be of interest to researchers in artificial intelligence who seek to produce adaptive behavior in virtual and mechanical agents (McDowell). The fine structure of behavior generated by the theory is similar to the fine structure of behavior generated by live organisms on similar schedules. This is evident from inspection of cumulative records, and from studies of interresponse time distributions in the form of log survivor plots (Kulubekova). A competing computational theory, proposed by Catania, does not fare as well as the evolutionary theory in describing steady-state behavior, although it produces some behavior with fine structure similar to that produced by live organisms (Berg). Detailed analyses show that the evolutionary theory in fact produces steady-state behavior that is consistent with the modern theory of matching, which has been proposed to replace the classic version of matching theory. Steady state behavior produced by the evolutionary theory shows systematic deviations from classic matching theory, but not from the modern theory (Caron). |
| |
| The Evolutionary Dynamics of Selection by Consequences: Theory and Data. |
| JACK J. MCDOWELL (Emory University) |
| Abstract: A computational theory of selection by consequences that implements Darwinian principles of selection, reproduction, and mutation with respect to a population of potential behaviors yields equilibrium behavior in single and concurrent schedules that is consistent with matching theory. In addition to its relevance for behavior analysis, this theory may be of interest to researchers in artificial intelligence who seek to produce adaptive behavior in virtual and mechanical agents. |
| |
| Fine Structure of Behavior Produced by the Evolutionary Theory. |
| SAULE KULUBEKOVA (Emory University) |
| Abstract: The cumulative records for random interval and random ratio schedules of reinforcement suggest that the fine structure of behavior generated by the computational model is similar to the fine structure of behavior generated by live organisms on similar schedules. The behavior of the computational model on RR schedules of reinforcement appears to be consistent with characteristics of VR (RR) responding in live organisms, such as the decreasing form of the response rate versus mean ratio function, ratio strain, and a boundary mean ratio beyond which responding ceases. Log survivor plots for the computational model did not show a sharp change in slope indicative of the “two limb” pattern of response bouts and pauses. Instead, the transition was gradual. Increasing the reinforcement rate increased the slope of the right-hand limb in log survivor plots, implying an increase in bout-initiation rate. These findings suggest that the fine structure of behavior generated by the theory is similar to the fine structure of behavior generated by live organisms. |
| |
| Comparison of the Evolutionary Theory to Catania’s Computational Model Based on the Operant Reserve. |
| JOHN PEDER BERG (Emory University) |
| Abstract: Skinner proposed a dynamic theory of behavior that conceptualized the function of reinforcement as setting up potential future responses. The total number of potential responses was the “reserve.” Responses emitted without reinforcement drew down the reserve while those with reinforcement increased the reserve. Catania implemented Skinner’s theory of the “operant reserve” in a computational model where the total reserve size was fixed but had an internal, variable level. The probability of response at each computational moment was proportional to the reserve level. All responses depleted the reserve. Behaviors immediately previous to a reinforcement event contributed to the reserve in decreasing amounts as determined by a reciprocal decay function. Catania qualitatively demonstrated correspondence between the operant reserve computational model and established molecular and molar behavioral phenomenon. The present study replicated the operant reserve computational model and quantitatively tested the equilibrium results for correspondence with the classic and modern theories of matching. Although some fine structure behavior was qualitatively similar to live organisms, the operant reserve model did not produce behavior consistent with matching theory. These results suggest that the operant reserve does not provide a plausible account of dynamic behavioral processes. |
| |
| Relationship of the Evolutionary Dynamics to the Classic and Modern Theories of Matching. |
| MARCIA LYNN CARON (Emory University) |
| Abstract: A computational theory of selection by consequences that implements Darwinian principles of selection, reproduction, and mutation with respect to a population of potential behaviors, yields equilibrium behavior in single and concurrent schedules that is consistent with matching theory. However, detailed analyses show that the evolutionary theory in fact produces steady-state behavior that is consistent with the modern, rather than the classic, theory of matching. The modern theory has been proposed to replace the classic version of the theory (as defined by Herrnstein’s original hyperbola) by allowing for bias and undermatching in separate bias and exponent parameters, respectively. Steady state behavior produced by the evolutionary theory shows systematic deviations (i.e., non-random residual profiles) from the classic theory of matching, but not from the modern theory. In addition, bias parameters obtained from fits to computational data vary with properties of the model believed to be related to the magnitude of reinforcement (i.e., the mean of the parental fitness function). Finally, undermatching appears to be an emergent property of the evolutionary theory, inasmuch as estimated exponents of about 0.8 are obtained from fits to the computational data, a value that is comparable to exponents obtained from live organisms. |
|
| |