Implicit adaptation is driven by direct policy updates rather than forward-model-based learning

The human motor system can rapidly adapt its motor output in response to errors, reducing errors in subsequent movements and thereby improving performance. It remains unclear, however, exactly what mechanism supports this learning. It has been proposed that the implicit adaptation of motor commands in response to errors occurs through updating an internal forward model which predicts the consequences of motor commands. This model can then be used to select appropriate motor commands that will lead to a desired outcome. Alternatively, however, it has been suggested that implicit adaptation might occur by using errors to directly update an underlying policy (often referred to as an inverse model). There is currently little evidence to distinguish between these two possibilities. Here, we exploit the fact that these two different learning architectures make opposing predictions about how people should behave under mirror-reversed visual feedback, and find that peoples9 behavior is consistent with direct policy learning, but not with forward-model-based learning.