Amygdala Acts as Decision Referee Rather Than Fear Alarm

You’re standing in front of an unfamiliar coffee machine. You’ve seen this type before, roughly — there’s a flashing button on the left. Do you press it because it was flashing, or because it was on the left? When reward arrives, your brain has a problem: it can’t tell which feature actually produced the result. So it runs both theories in parallel, quietly, and waits to find out which one is more reliable.

Somewhere inside that process, tucked deep in the temporal lobe, a pair of almond-shaped clusters of neurons is doing something neuroscientists have largely overlooked. The amygdala — long cast as the brain’s smoke detector, the trigger of racing hearts and sweaty palms — turns out to be running a far more subtle operation. According to a study published in Nature Communications, it acts as an arbitrator between competing learning systems, tilting the brain toward whichever model of the world is actually working.

“There is nothing really primitive in the brain, even when you talk about this area,” says Alireza Soltani, associate professor of psychological and brain sciences at Dartmouth College and the study’s senior author. “People have labeled the amygdala as an emotional fear system,” he says, but the new work points toward a more complex function — one that only becomes visible when things are genuinely uncertain.

Soltani and PhD candidate Jae Hyung Woo built their investigation around a deceptively simple question: how does the brain choose between different ways of learning? When we navigate reward and punishment, we don’t rely on a single strategy. Action-based learning tracks what you did — press the left button, get coffee. Stimulus-based learning tracks what you saw — the flashing light predicted success. Both run simultaneously, and both accumulate evidence. The question is which one the brain trusts more at any given moment.

“The key distinction,” says Soltani, “is whether learning should be tied to a motor action or the identity of the stimulus.” Stimulus-based learning is more flexible — you can evaluate options before committing to movement. But action-based learning is faster, more automatic, anchored to the body. Neither is better in principle; each works better depending on the task. What the brain needs, then, is a mechanism to arbitrate between them.

To find that mechanism, the team compared three groups of monkeys performing a probabilistic learning task designed to be genuinely ambiguous. Rewards shifted without warning. The animals didn’t know, at the start of each new block of trials, whether rewards were tied to which object they picked or where they picked it — a feature of many real-world learning situations, where the rules are unknown until they’re inferred. Some monkeys had intact brains. Others had lesions to the amygdala. A third group had lesions to the ventral striatum, a structure involved in reward-based learning.

The results split cleanly. Monkeys with ventral striatum damage showed what you’d predict: impaired stimulus-based learning, a bias toward action. But their arbitration process — the mechanism weighting one system against the other — remained largely intact. They could still update which model the brain was relying on; they just had a weaker signal in one channel.

Amygdala-lesioned monkeys were different in a telling way. Their arbitration didn’t just skew — it flattened. The brain could no longer efficiently update which learning system was more reliable. It defaulted toward action-based learning from the start and failed to correct course as evidence accumulated. Behaviour became rigid, even in situations where the stimulus-based system would have served better. “A healthy amygdala promotes exploration between alternative models,” says Soltani, “and as a result, can make you choose something you wouldn’t otherwise choose, and you can learn from that.”

This goes some way to resolving a puzzle that has nagged at amygdala research for decades. Earlier studies produced a confusing picture: damage to the structure sometimes impaired learning, sometimes improved it. A standard single-strategy model couldn’t account for the pattern. Woo and Soltani’s computational framework suggests why: the effect depends on which learning system the task engages, and how much evidence has built up before the amygdala’s arbitration role kicks in. In some conditions, leaning harder on the action-based system actually helps — particularly when the stimulus system has already latched onto the wrong answer and needs dislodging. The amygdala’s normal role is to manage that tension. Without it, the brain gets stuck in whatever groove it started in.

“Historically, the amygdala has been studied from the perspective of fear learning, and it has been generalized to reward learning,” says Woo. “Our main hypothesis was that it must have other functions given its extensive connections to the rest of the brain.” Those connections run to the orbitofrontal cortex, the prefrontal cortex, the ventral striatum — regions involved in planning, valuing, deciding. They suggest a structure with a hand in many processes, not a specialist in one.

The research team is now recording neural activity in the prefrontal cortex during these tasks, and running experiments with rats to trace specific pathways between the amygdala and prefrontal regions. The hope is to move from the computational description — which model is being weighted, and when — to the circuit-level detail of how that weighting actually happens.

The practical implications are already taking shape. The team points to phobias as one example. Someone afraid of spiders tends to get locked into stimulus-based learning: the spider is the problem, and the brain’s response revolves entirely around that association. Avoidance entrenches the pattern. “The fear is tied to the stimulus,” says Soltani, “making the response rigid and difficult to override.” His suggestion — shifting attention to an action-based approach, placing a cup over the spider rather than trying to rationalise the threat away — is an attempt to engage the other learning channel. Whether that works clinically remains to be tested, but the logic follows from what the arbitration model predicts.

What the study quietly establishes is that the brain’s architecture for decision-making is more layered than the fear-circuit narrative has tended to suggest. “We’re not saying the amygdala isn’t about fear,” says Woo. “We’re reframing its role as a mediator between multiple learning systems.” A mediator doesn’t sound dramatic. But in a brain that’s perpetually weighing competing theories about how the world works, a reliable referee might be among the most important things you have.

Study link: https://www.nature.com/articles/s41467-025-66745-1