Human sensorimotor adaptation critically depends on the ability to map sensory inputs onto motor outputs. While this process was once thought to rely primarily on cerebellar-dependent implicit recalibration, recent work has revealed that explicit, cognitive strategies, relying on working memory and executive function, play a substantial role. However, it remains unclear whether explicit strategies can scale to support learning of complex sensorimotor mappings. Here, we parametrically tested the capacity of explicit strategies to solve a complex visuomotor rotation task by varying the number of target- rotation pairings that participants had to acquire. In experiment 1, participants were tasked with learning four visuomotor mappings - well within working memory capacity, based on prior studies. We found that participants achieved near-perfect compensation, which was best explained by the retrieval of stored target-rotation associations, rather than by computationally demanding algorithmic strategies. In experiment 2, in an attempt to push beyond working memory capacity, participants were tasked with learning eight mappings. Unlike in experiment 1, here we found that participants failed to fully compensate for the rotations, reaching asymptotic performance of only ~50%, despite evidence of continued strategic engagement. This performance limit was fully predicted by a parameter-free working-memory model where performance is a mixture of a fixed number of stored target-rotation associations and random guessing. These findings reveal a cognitive "bandwidth limit" on the effectiveness of strategies for sensorimotor adaptation: when task complexity exceeds this limit, adaptation plateaus, defining a fundamental constraint on how far higher-order cognition can go to support learning.
Bejjanki et al. (Mon,) studied this question.