Human hippocampal ripples prioritise model-based learning

Humans often learn optimally, inferring the value of options they have never directly experienced by leveraging internal models of the world, a process known as model-based learning. Yet how the brain decides which unexperienced options to update first remains unclear. Here we recorded intracranial EEG from 34 epilepsy patients performing a reinforcement-learning task that required using task structure to infer the value of unvisited (non-local) paths. After each reward, brief hippocampal “ripple” events signalled which indirect experience was most valuable, thereby encoding that path’s priority. Longer ripples carried the strongest priority signals, allowing the brain to update high-value, unvisited options first and thereby optimise learning. Ripple events also coincided with selective cortical reactivation of these high-priority paths, consistent with prioritised replay. Crucially, when hippocampal ripples were precisely synchronised with activity in the lateral frontopolar cortex (LFPC), individuals showed greater model-based learning: stronger ripple–LFPC coupling predicted more effective use of task structure and more accurate learning of non-local values. Our findings establish hippocampal ripple-centred prefrontal coordination as a fundamental mechanism that prioritises valuable experiences for model-based learning, explaining how the human brain learns so efficiently.