Mice trained to associate a sound with sugar water learned the association with fewer repetitions when the trials were spaced five to ten minutes apart than when they were spaced thirty to sixty seconds apart. The dopamine response — the neural signature of learned prediction — emerged faster from infrequent exposure. The brain needed fewer rare events than frequent events to form the same association.
The finding inverts the intuition that practice makes perfect through repetition. More exposures should produce faster learning, if learning is a matter of accumulation — each trial adding a small increment to the strength of the association. But accumulation is not the only learning mechanism. The brain also learns by extracting causal structure: identifying which events predict which outcomes. Causal extraction benefits from spacing because spacing reduces interference. When trials are close together, the neural representation of one trial bleeds into the next, making it harder to isolate the predictive relationship. When trials are far apart, each trial is a clean signal against a quiet background.
The most striking result involved intermittent reinforcement. When the sound was played every sixty seconds but the sugar water was delivered only ten percent of the time, the mice needed far fewer actual rewards before dopamine began responding to the sound alone — regardless of whether that specific trial was rewarded. The rarity of the reward made each occurrence more informative. A reward that happens every time is expected; its occurrence says nothing new. A reward that happens ten percent of the time is surprising; its occurrence updates the predictive model substantially.
This is the mathematical structure of information theory applied to neural learning. The information content of an event is inversely proportional to its probability. A common event carries little information. A rare event carries a lot. If the dopamine system is tracking information content rather than raw frequency, then rare rewards should produce faster learning per unit of reward — which is exactly what the study found. The brain is not counting. It is estimating.
The practical implications extend beyond neuroscience. Educational research has long known that spaced repetition outperforms massed practice for long-term retention. The Burke study suggests a mechanism: spacing does not just prevent fatigue or boredom. It changes the signal-to-noise ratio of each learning event. A study session after a long gap occurs against a background of partial forgetting, which means the brain can more clearly distinguish what is being reinforced from what is already known. The forgetting is not a cost of spacing. It is the mechanism by which spacing works.
The connection to dopamine is specific. Dopamine neurons encode prediction error — the difference between expected and observed outcomes. An expected reward produces little dopamine. An unexpected reward produces a lot. Spacing increases the surprise of each reward because the prediction has had time to decay between events. The dopamine burst is larger, the learning update is stronger, and the association forms from fewer total events. The system learns most efficiently when it is slightly uncertain.