Related Experiment Videos
A selectionist approach to reinforcement
J W Donahoe1, J E Burgos, D C Palmer
1Department of Psychology, University of Massachusetts, Amherst 01003.
Journal of the Experimental Analysis of Behavior
|July 1, 1993
Summary
A new reinforcement principle, modeled in neural networks, explains complex behaviors like learning and stimulus control. This biologically plausible model simulates respondent and operant conditioning phenomena.
Area of Science:
- Neuroscience
- Behavioral Science
- Computational Neuroscience
Background:
- Traditional models of reinforcement often treat behavior and neural processes separately.
- Understanding the neural basis of reinforcement is crucial for explaining complex adaptive behaviors.
Purpose of the Study:
- To introduce a unified principle of reinforcement integrating behavioral and neuroscientific data.
- To explore the implications of this principle using computer simulations of adaptive neural networks.
Main Methods:
- Development of a biologically plausible neural network model.
- Implementation of a single reinforcement principle within the network.
- Computer simulations to observe emergent network behaviors.
Main Results:
- The model successfully simulated key phenomena of respondent and operant conditioning, including acquisition, extinction, and stimulus control (e.g., blocking, discrimination).
- Simulated networks demonstrated context-dependent behavior guidance, consistent with behavior-analytic principles.
- Reinforcement acting on network connectivity produced complex behaviors from simple processes.
Conclusions:
- A single, biologically plausible reinforcement principle can account for a wide range of behavioral phenomena.
- Adaptive neural networks offer a powerful framework for simulating complex behavior and understanding its neural underpinnings.
- This selectionist approach provides a promising avenue for simulating the complex behavior of living organisms.