Understanding the development of reward learning through the lens of meta-learning

A mechanistic account of how our environments shape how we learn is critical for understanding individual differences in behavior. In this talk, I will propose that specific statistics of experienced environments shape reward learning across multiple nested timescales via a meta-learning process, focusing on the example of environmental controllability. The controllability of positive or negative experiences has long been recognized as a critical factor determining their impact on subsequent behavior. Building on an extensive body of work in animal models, I will argue that humans use estimates of environmental controllability to calibrate motivated behavior along a continuum ranging from proactive (‘what can I do in this environment?’) to reactive ('what will this environment do to me?’). I will present studies suggesting that the controllability of motivationally significant outcomes not only modulates an individual's balance between proactive and reactive reward learning strategies over the short timescales of experimental tasks, but also informs individual differences in reward-guided behavior over the course of development. I will discuss the computational processes and neural mechanisms that may underpin this meta-learning process, and how such environmental adaptation may modulate individual vulnerability to psychopathology.

Speaker: Catherine Alexandra Hartley, New York University