1. ¹Psychology Department, McMaster University, Hamilton, Ontario, Canada
2. ²Centre for Addiction and Mental Health, Toronto, Ontario, Canada
3. ³Canada Research Chair and Professor of Psychiatry, University of Toronto, Vice President Research, Centre for Addiction and Mental Health, Toronto, Ontario, Canada

Correspondence: Dr S Kapur, Canada Research Chair and Professor of Psychiatry, University of Toronto, Vice President Research, Centre for Addiction and Mental Health, 250 College Street, Toronto, Ontario, M5T 1R8 Canada. Tel: +1 416 979 6890; Fax: +1 416 260 4206; E-mail: shitij_kapur@camh.net

Received 30 October 2003; Accepted 6 January 2004; Published online 3 March 2004.

Abstract

The selective ability of antipsychotic drugs (APDs) to attenuate conditioned avoidance responding (CAR) has been recognized for over 50 years. However, most efforts to account for this finding have been either neurochemically oriented (focusing on the neuromodulator dopamine) or behavioral, with little effort invested in uniting the two within a computational model. In this paper we propose a computational model, based on concepts from formal reinforcement learning theory, which accounts for the basic finding that noncataleptic doses of APDs disrupt avoidance without disrupting escape. The model formally separates out sensory, motor, and reward processes, and makes novel predictions pertaining to the dose- and time-dependent effects of APDs on response latencies—predictions which we verified in experimental studies using four different APDs (haloperidol, chlorpromazine, risperidone, and clozapine). The APD action in this model is most consistent with an effect on 'expected future reward'—an idea closely linked to motivational drives and consistent with several leading theories of dopamine action.