Time is a perishable resource that cannot be reallocated once it has elapsed. As a result, time allocation decisions must often be adjusted on the basis of information that becomes available during the process of allocation itself. We show that adaptive rules of this kind will not generally result in optimal allocations when the relationship between the optimal allocation and the time budget exhibits discontinuities, as is the case with logistic learning curves. Under such circumstances, self-guided study is likely to perform poorly relative to appropriately guided learning.

Using a new agent-based model that mimics the learning process in hand-reaching behavior of individual mice, we show that mouse hand preference is probabilistic, dependent on the environment and prior learning. We quantify the learning capabilities of three inbred strains and show that population distributions of hand preference emerge from the properties of individual mice. The model informs our understanding of gene–environment interactions because it accommodates genotypic differences in learning and memory abilities, and environmental biases. We tuned each strain’s model to match their experimental hand-preference distributions in unbiased worlds and, by comparing simulations and experiments, identified and quantified a constitutive left-bias in hand preference of one strain. The models, tuned for unbiased worlds, match experimental measures in left- and right-biased worlds and in biased worlds after previous training. New measures quantitatively assess this matching, revealing that two strains, previously considered non-learners of hand preference, actually have significant learning ability and we confirm this with new experiments. Model mice match the kinetics of hand-preference learning of one strain and predict the limits of learning. We conclude that genetically evolved hand-preference behavior in mice is inherently probabilistic to provide robustness and allow constant adaptability to ever-changing environments.