Individualised dynamic internal representations from response times

Internal models capture the regularities of the environment and are central to understanding how humans adapt to environmental statistics. In general, the correct internal model is unknown to observers, instead approximate and transient ones are recruited. However, experimenters assume an ideal observer model, which captures stimulus structure but ignores the diverging hypotheses that humans form during learning. We combine non-parametric Bayesian methods and probabilistic programming to infer rich and dynamic individualised internal models from response times in an implicit visuomotor sequence learning task. We identify two contributors to the internal model: the ideal observer model and a Markov model capturing only immediate temporal dependencies between observations. Individual learning curves revealed internal models initially dominated by the Markov model, which was later traded-off with the ideal observer model. Thus, our results reveal a structured inductive bias that varies across individuals both in strength and persistence but is consistent in overall structure.