Predictive Representations: Building Blocks of Intelligence

Wilka Carvalho; Momchil S Tomov; William de Cothi; Caswell Barry; Samuel J Gershman

doi:10.1162/neco_a_01705

Predictive Representations: Building Blocks of Intelligence

Neural Comput. 2024 Aug 30:1-74. doi: 10.1162/neco_a_01705. Online ahead of print.

Authors

Wilka Carvalho¹, Momchil S Tomov^{2

3}, William de Cothi⁴, Caswell Barry⁵, Samuel J Gershman^{6

7}

Affiliations

¹ Kempner Institute for the Study of Intelligence, Harvard University, Cambridge, MA 02134, U.S.A. wcarvalho92@gmail.com.
² Department of Psychology and Center for Brain Science, Harvard University, Cambridge, MA 02134, U.S.A.
³ Motional AD LLC, Boston, MA 02210, U.S.A. tomov90@gmail.com.
⁴ Department of Cell and Developmental Biology, University College London, London WC1E 7JE, U.K. w.decothi@ucl.ac.uk.
⁵ Department of Cell and Developmental Biology, University College London, London WC1E 7JE, U.K. caswell.barry@ucl.ac.uk.
⁶ Kempner Institute for the Study of Intelligence, and Department of Psychology and Center for Brain Science, Harvard University, Cambridge, MA 02134, U.S.A.
⁷ Center for Brains, Minds, and Machines, MIT, Cambridge, MA 02139, U.S.A. gershman@fas.harvard.edu.

PMID: 39212963
DOI: 10.1162/neco_a_01705

Abstract

Adaptive behavior often requires predicting future events. The theory of reinforcement learning prescribes what kinds of predictive representations are useful and how to compute them. This review integrates these theoretical ideas with work on cognition and neuroscience. We pay special attention to the successor representation and its generalizations, which have been widely applied as both engineering tools and models of brain function. This convergence suggests that particular kinds of predictive representations may function as versatile building blocks of intelligence.