Pavlovian Impulsivity: Connectionist Predictions, with Implications for Addiction

Impulsivity, defined as the preference for a small reward after a short delay over a large reward after a long delay, has been widely studied as possibly underlying addictive behavior. Animal studies have focused on operant conditioning, where reinforcement is response dependent. In this paper, we use an extant neural-network model of conditioning to predict a new, empirically testable phenomenon we call “Pavlovian impulsivity,” defined as impulsive behavior that emerges from Pavlovian (response-independent) reinforcement. A computer simulation was run using a feedforward neural network with inputs to simulate two discrete cues (X, Y), a continuously present context (C), and a primary reinforcer that varied in magnitude (small vs. large) and delay (short vs. long). The network also had outputs to simulate two noncompetitive conditioned responses that simulated emitted responding in not being unconditionally elicited by the reinforcer. The network first received X-reward pairings with a small reinforcer after a short interstimulus interval (ISI), randomly intermixed with Y-reward pairings with large reinforcer after a long ISI, all in C. Rewards did not depend on the output activations, to simulate response-independent reinforcement. Then, the network received choice trials of the compound XY in C. Pavlovian impulsivity was simulated as a preference for X. Implications for a better understanding impulsive behavior are discussed.