In noisy, dynamic environments, organisms must distinguish genuine change (e.g., the movement of prey) from noise (e.g., the rustling of leaves). Expectations should be updated only when the organism believes genuine change has occurred. Although individual variables can be highly unreliable, organisms can take advantage of the fact that changes tend to be correlated (e.g., movement of prey will tend to produce changes in both visual and olfactory modalities). Thus, observing a change in one variable provides information about the rate of change for other variables. We call this the penumbra of learning. At the neural level, the penumbra of learning may offer an explanation for why strong plasticity in one synapse can rescue weak plasticity at another (synaptic tagging and capture). At the behavioral level, it has been observed that weak learning of one task can be rescued by novelty exposure before or after the learning task. Here, using a simple number prediction task, we provide direct behavioral support for the penumbra of learning in humans, and show that it can be accounted for by a normative computational theory of learning.