PREDICTING THE WORLD THROUGH EXPERIENCE AND SENSORY EVIDENCE

Our experience of reality is not a direct reflection of external events, but an active construction shaped by prior beliefs and sensory input. The brain functions as a predictive engine, continuously updating internal models to reconcile expectations with incoming evidence. When this process is disrupted, the boundary between perception and reality can blur, leading to maladaptive beliefs and distorted percepts. Predictive-Coding theories provide a unifying framework, proposing that perception and decision-making emerge from interactions between bottom-up-sensory-evidence and top-down-expectations via prediction-errors. Disruptions in this balance are thought to underlie several psychiatric disorders, yet studying these mechanisms in humans remains challenging, particularly at the circuit-level. To address this, we developed and validated a novel three-arm-bandit-task for wild-type-mice, assessing how prior beliefs and sensory evidence are integrated under changing task demands and probabilistic reward structures. Behavioral results show that mice use sensory cues to infer latent task states, adapt to block transitions, and update strategies following unexpected sensory feedback. To investigate the neural basis of these processes, we record neuronal activity in the medial prefrontal cortex and striatum, regions implicated in state representation, inference under uncertainty, and reward prediction. In parallel, we apply this task to a mouse model of cognitive impairment (Df(16)A+/–). Preliminary data indicate delayed learning and increased sensitivity to sensory changes, yet similar performance levels to controls, consistent with behavioral findings in patients with psychosis. Finally, we aim to adapt this task for human participants using a shared predictive coding framework, enabling cross-species comparison and strengthening translational relevance.