Premium
CHANGE DETECTION, MULTIPLE CONTROLLERS, AND DYNAMIC ENVIRONMENTS: INSIGHTS FROM THE BRAIN
Author(s) -
Pearson John M.,
Platt Michael L.
Publication year - 2013
Publication title -
journal of the experimental analysis of behavior
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.75
H-Index - 61
eISSN - 1938-3711
pISSN - 0022-5002
DOI - 10.1002/jeab.5
Subject(s) - reinforcement learning , computer science , reinforcement , artificial intelligence , control (management) , key (lock) , contrast (vision) , anterior cingulate cortex , temporal difference learning , operant conditioning , machine learning , psychology , neuroscience , cognition , social psychology , computer security
Foundational studies in decision making focused on behavior as the most accessible and reliable data on which to build theories of choice. More recent work, however, has incorporated neural data to provide insights unavailable from behavior alone. Among other contributions, these studies have validated reinforcement learning models by demonstrating neural signals posited on the basis of behavioral work in classical and operant conditioning. In such models, the values of actions or options are updated incrementally based on the difference between expectations and outcomes, resulting in the gradual acquisition of stable behavior. By contrast, natural environments are often dynamic, including sudden, unsignaled shifts in reinforcement contingencies. Such rapid changes may necessitate frequent shifts in behavioral mode, requiring dynamic sensitivity to environmental changes. Recently, we proposed a model in which cingulate cortex plays a key role in detecting behaviorally relevant environmental changes and facilitating the update of multiple behavioral strategies. Here, we connect this framework to a model developed to handle the analogous problem in motor control. We offer a tentative dictionary of control signals in terms of brain structures and highlight key differences between motor and decision systems that may be important in evaluating the model.