Topics in Cognitive Science

Original Article

Free Access

A Contrast‐Based Computational Model of Surprise and Its Applications

Corresponding Author

CISUC, Department of Informatics Engineering, University of Coimbra

Correspondence should be sent to Luis Macedo, CISUC, Department of Informatics Engineering, University of Coimbra, Pólo II ‐ Pinhal de Marrocos 3030‐290 Coimbra, Portugal. E‐mail:

macedo@dei.uc.pt

Search for more papers by this author

Amílcar Cardoso

CISUC, Department of Informatics Engineering, University of Coimbra

Search for more papers by this author

Luis Macedo

Corresponding Author

E-mail address: macedo@dei.uc.pt

CISUC, Department of Informatics Engineering, University of Coimbra

Correspondence should be sent to Luis Macedo, CISUC, Department of Informatics Engineering, University of Coimbra, Pólo II ‐ Pinhal de Marrocos 3030‐290 Coimbra, Portugal. E‐mail:

macedo@dei.uc.pt

Search for more papers by this author

Amílcar Cardoso

CISUC, Department of Informatics Engineering, University of Coimbra

Search for more papers by this author

First published: 19 November 2017

https://doi.org/10.1111/tops.12310

Cited by: 1

This article is part of the topic “The Ubiquity of Surprise: Developments in Theory, Converging Evidence, and Implications for Cognition,” Edward Munnich, Meadhbh Foster and Mark Keane (Topic Editors). For a full listing of topic papers, see http://onlinelibrary.wiley.com/journal/10.1111/(ISSN)1756-8765/earlyview.

Abstract

We review our work on a contrast‐based computational model of surprise and its applications. The review is contextualized within related research from psychology, philosophy, and particularly artificial intelligence. Influenced by psychological theories of surprise, the model assumes that surprise‐eliciting events initiate a series of cognitive processes that begin with the appraisal of the event as unexpected, continue with the interruption of ongoing activity and the focusing of attention on the unexpected event, and culminate in the analysis and evaluation of the event and the revision of beliefs. It is assumed that the intensity of surprise elicited by an event is a nonlinear function of the difference or contrast between the subjective probability of the event and that of the most probable alternative event (which is usually the expected event); and that the agent's behavior is partly controlled by actual and anticipated surprise. We describe applications of artificial agents that incorporate the proposed surprise model in three domains: the exploration of unknown environments, creativity, and intelligent transportation systems. These applications demonstrate the importance of surprise for decision making, active learning, creative reasoning, and selective attention.