Interaction between Model-based and Model-free Mechanisms in Motor Learning

Motor learning can be driven by habitual, model-free processes, or strategic, model-based processes—depending on the magnitude and context of movement errors. Although small and large errors are known to engage distinct motor learning mechanisms—model-free (implicit) or model-based (explicit), respectively—it remains unclear whether successfully deploying one mechanism might impede the other in subsequent learning tasks. Here, we investigated how prior engagement of a particular mechanism biases future adaptations, even when the new task context typically favours the alternative strategy. Across three experiments (N=82), participants performed reaching movements to targets that either remained fixed or “jumped” mid-movement by small (15°) or large (30°, 45°, or 60°) angles. When first exposed to small errors, participants exhibited persistent aftereffects and stable reaction times (RTs), hallmarks of a model-free process. Surprisingly, despite larger error magnitudes later, these participants continued to show robust aftereffects and did not elevate RTs–indicating a carryover of model-free learning. Conversely, participants who initially experienced large errors showed minimal aftereffects and flexible RT modulation consistent with model-based strategies; this bias persisted in later sessions with smaller errors, leading to reduced habitual aftereffects. Notably, inserting a washout phase to reset baseline performance did not abolish these mechanistic biases, highlighting that the initial engagement of either model-free or model-based processes leaves a durable imprint on subsequent adaptations. These findings demonstrate that motor learning is shaped not only by ongoing task demands but also by prior learning history, with implications for designing interventions and training protocols in motor rehabilitation.