The Neural Substrates of Deterministic Decision-making

[摘要] When making a decision, we draw upon multiple mnemonic resources to inform our behavior and to ideally produce a good outcome. Multiple memory systems guide this process, including a medial temporal lobe (MTL) system and a striatal system. The MTL provides episodic details about specific instances of prior experience, whereas the striatum provides a prediction about possible outcomes based upon a fusion of many prior experiences. While both of these systems are assumed to support decision behavior, extricating their discrete contributions has been challenging. Using neuroimaging and computational reinforcement learning, this study investigated the extent to which the MTL and striatal systems are co-active during single-exposure learning and how these systems each support subsequent behavior. This was done in the context of a single-exposure deterministic decision-making task that separated encoding processes from subsequent decision-making processes. Human subjects learned to associate words with monetary feedback in a single decision experience. They then used that information to make better choices in a subsequent round without feedback. Activity in MTL regions predicted episodic memory accuracy and correlated with subsequent decision accuracy and response times. Additionally, the MTL supported a model-based reinforcement learning process wherein initial decision experiences were used to build a model of the environment that was then used to prospectively formulate future decision outcome predictions. Activity in striatal regions also correlated with subsequent decision accuracy and response times, but did not relate to memory accuracy. The striatum supported a model-free reinforcement learning process wherein predictions about decision outcomes were generated from a retrospective accumulation of prior decision experiences. Together, these results implicate both the MTL and striatum as essential substrates to single-exposure learning, but underscore that these systems operate in fundamentally different ways. The MTL is associated with prospective learning, wherein single instances of prior experience can be leveraged to inform subsequent choice. The striatum, in contrast, is associated with retrospective learning, wherein a history of experience is required to build reliable predictions about subsequent choices. In combination, the MTL system seems to support decision behavior until the striatal system has had enough experience to refine predictions about outcomes.

[发布机构] the University of Pittsburgh