A single peer-reviewed study recorded individual neurons while seven people with treatment-resistant epilepsy played an asteroid-avoidance game. Activity in the claustrum, a thin structure deep in the brain, varied with modeled uncertainty and prediction errors. The recordings show information associated with learning, while clinical electrode placement and the small sample limit broader conclusions.
A single peer-reviewed study examined the claustrum, a thin structure deep inside the brain, during learning from aversive outcomes. Of 110 claustrum neurons evaluated, 78 responded to task events. Signals associated with modeled uncertainty appeared in 31 neurons; 25 carried activity linked to prediction errors, the difference between expected and actual outcomes. Some model variables changed together, limiting separation of the signals.[1]
Seven participants, aged 18 to 61, controlled a spaceship while avoiding asteroids. Safe zones at the top and bottom of the screen offered changing chances of success, requiring participants to learn from earlier outcomes. A behavioral model estimated uncertainty and prediction errors from the task. Participants had treatment-resistant epilepsy and received electrodes for clinical seizure localization. That medical need determined electrode placement rather than a design sampling the general population.[1]
The researchers recorded neurons without stimulating or silencing them. The measurements identify activity associated with the task, while leaving causal effects of changing that activity untested. Epilepsy, medication and the small clinical sample also limit application of the findings to other people.[1]
The comparison included 80 neurons in the anterior cingulate cortex and 75 in the amygdala, two other brain regions. The timing of uncertainty-related activity differed between the claustrum and anterior cingulate cortex. Claustrum signals were especially apparent during active avoidance, and responses to crashes were associated with aversive feedback. The ability of a model to fit behavior leaves open which computations the brain actually performs.[1]
