Amygdala mediates interbrain alignment in social dilemmas

Summary

The study identified an amygdala-mediated 'simulation-to-decision' cascade in a real-time, iterated prisoner's dilemma experiment using simultaneous intracranial EEG data from pairs of epilepsy patients. In this cascade, inter-brain spectral alignment (IBRA) emerges selectively only during mutual cooperation and follows model-based forward predictions of the partner's intent. This prediction is then linked to the individual's own decision via θ–γ coupling.

Why it matters

Cooperation has long been modeled as an internal, isolated computation. This study provides direct intracranial evidence that two interacting brains enter a measurable, frequency-based alignment state. This state is both necessary and predictive for cooperative outcomes. If the findings are replicated, the social decision-making process will be reframed as a genuinely dyadic neural process. Additionally, it will offer a mechanistic explanation for why some individuals coordinate under uncertainty while others avoid cooperation.

Highlights

• Early IBRA (θ suppression) encodes rapid, history-sensitive partner tracking. Late IBRA (low-frequency suppression + γ increase) supports committed action.
• Conditional cooperators rely on partner-tracking signals. Unconditional cooperators rely on a tendency-based tolerance for uncertainty; this shifts them toward predictive alignment.
• The amygdala serves a central function. Low-frequency suppression follows forward predictions of partner cooperation. θ–γ coupling bridges this prediction to the individual's own choice.

Critical take

Because the sample is a small clinical cohort of epilepsy patients, generalizability to the healthy population is uncertain. Furthermore, the iEEG design is inherently correlational. The causal direction (does amygdala alignment *drive* cooperation or merely *accompany* it?) has not yet been tested. Grand claims like 'redefining cooperation' also go beyond what a single-task, two-person paradigm can establish.