For a century, the fundamental model of learning has been repetition. Pavlov rang the bell, delivered the food, rang the bell, delivered the food. More pairings, stronger associations. The framework is so embedded that it extends beyond neuroscience into education, training, and machine learning: more examples, better performance. The count is what matters.
Burke, Taylor, Jeong, and colleagues at UCSF ran the obvious experiment that hadn't been done (Nature Neuroscience, 2026). They trained mice to associate a sound with sugar water but varied the spacing between trials. Mice with trials 5-10 minutes apart needed far fewer repetitions before their dopamine systems began firing at the cue — the neurochemical signature of learning. More striking: mice that received the reward only 10% of the time but with well-spaced cues learned faster than mice drilled with twenty times more trials over the same period.
The learning system doesn't count repetitions. It measures intervals.
The mechanism inverts the standard picture. Dense repetitions don't reinforce the association — they dilute the signal. Each rapid pairing adds a data point in a context where the previous data point hasn't been processed. The dopamine system needs time between events to update its model. Without that time, additional trials are noise, not signal. The mice that received 10% reward rates with spacing didn't learn from fewer rewards despite the rarity. They learned from fewer rewards because of the rarity. Each event landed in a cleared workspace.
As Namboodiri put it: “Associative learning is less 'practice makes perfect' and more 'timing is everything.'”
The distinction matters because repetition and spacing aren't two axes of the same optimization — they're structurally different. Repetition optimizes for total exposure. Spacing optimizes for per-event clarity. When the system is information-limited (each event carries the same amount of signal regardless of timing), repetition wins. When the system is processing-limited (each event needs integration time before the next one lands), spacing wins. Pavlov assumed the first. The brain operates under the second.
The general form: for any system that needs time to integrate information, rarer events produce stronger updates than frequent ones, and the optimal event rate is set by processing speed, not by the amount of information available. More is not better when the bottleneck is absorption.