Insight, priors, and multidimensional representation in sensorimotor adaptation

Sensorimotor adaptation can be understood as hierarchical inference over the causes of sensorimotor error and over the appropriate response. An agent must infer whether an error is relevant to their goals, what changes caused that error, and how behaviour should be adapted. Adaptation research has relied on simplifying abstractions that make behaviour tractable to measure and model, but these abstractions may obscure the processes that underlie human behaviour. This thesis attempts to alleviate that obscuration by developing analyses and models that ascribe greater nuance and richer representational capacity to the adaptation system. Chapter 2 examines the ubiquitous practice of analysing group-average behaviour. This abstraction obscures the fact that individual strategic re-aiming behaviour is qualitatively distinct from the group average. Rather than the smooth, incremental average behaviour, individuals perseverate in baseline behaviour before a single-trial shift to good performance. Chapter 3 examines strategic adaptation after this initial “Aha!” moment. Existing accounts of adaptation subscribe to convenient abstractions which ignore the more nuanced properties of the environment. The proposed novel model, combined with ecologically motivated priors, explain sign-flip errors, strategic savings, and individual differences in learning complex target-specific perturbations as predictable consequences of rational inference. Chapter 4 examines whether the study of implicit adaptation has similarly been obscured by adopting inappropriate abstractions. Implicit adaptation appeared to forego such simplifying abstractions. A more nuanced model of implicit adaptation captures human behaviour well, whereas previous models do not. These findings suggest that moving beyond convenient abstractions is necessary for understanding adaptation.