|
Implementing Punishment in an Evolutionary Theory of Behavior Dynamics |
Friday, September 2, 2022 |
10:30 AM–10:55 AM |
Meeting Level 2; Wicklow Hall 2A |
Area: PCH |
Chair: Jack J McDowell (Emory University) |
|
Implementing Punishment in an
Evolutionary Theory of Behavior Dynamics |
Domain: Theory |
JACK J MCDOWELL (Emory University) |
|
Abstract: A comprehensive theory of adaptive behavior is a desirable goal for a science of behavior. The evolutionary theory of behavior dynamics is one candidate for such a theory. It is a complexity theory that instantiates the Darwinian principles of selection, reproduction, and mutation in a genetic algorithm. The algorithm is used to animate artificial organisms that behave continuously in time and can be placed in any experimental environment. This presentation explains how punishment may be implemented in the theory. A key feature of this implementation is that when punishment is superimposed on reinforced responding, the suppressive effect of punishment depends on the amount of reinforcement generated by the target response. The suppressive effect is greater when less reinforcement is generated by the target response than when more is generated. This is a form of reinforcement loss aversion, or “fear of missing out,” as it is sometimes referred to colloquially. Relevant empirical evidence from experiments with live organisms is reviewed, including data from studies that superimpose punishment on concurrent and single alternative schedules. Findings from these studies support this implementation of punishment in the evolutionary theory. |
|
CANCELLED: Approximating McDowell’s Evolutionary Theory of Behavior Dynamics with Neural Networks |
Domain: Theory |
STEVEN J RILEY (Department of Psychology, Emory University) |
|
Abstract: McDowell’s evolutionary theory of behavior dynamics (ETBD) has successfully replicated the patterns of choice found in live animals under schedules of VI and VR reinforcement, both on molar and molecular scales. However, it bears little resemblance to the workings of animal bodies, and so lacks face validity. This presentation argues that the ETBD approximates a neural reinforcement process. First, it identifies an isomorphism between the space of outputs of the theory with the space of states in a neural network known as a restricted Boltzmann machine. Next, a “relaxation” of this network toward biological realism is presented, with real-valued connection weights that change like biological synapses. Both networks respond to reinforcement and produce sequences of outputs that replicate the major findings of the ETBD. This link from ETBD to biologically inspired networks gives a rigorous foundation to the selectionist metaphor of reinforcement. |
|
|
|