The Amygdala Is Not a Fear Alarm - It Referees Between Competing Learning Strategies
The amygdala has a reputation problem. Decades of fear-conditioning research branded it as the brain's primitive alarm bell - the structure that makes you flinch at a spider, freeze in a tight space, or break into a sweat at the edge of a balcony. That framing has been so influential it has shaped how clinicians think about anxiety disorders and how neuroscientists design their experiments.
A study published in Nature Communications by a Dartmouth research team argues that this framing is incomplete. The amygdala, the authors report, is doing something more sophisticated than reflexive threat detection. It actively mediates between two fundamentally different strategies the brain uses to learn and make decisions - and when it is damaged, that mediation breaks down in ways that reveal its true function.
Action-based versus stimulus-based learning
The conceptual starting point is a distinction that cognitive neuroscientists have studied for years but that has rarely been linked to the amygdala. Consider approaching an unfamiliar coffee machine. You might use action-based learning - pressing the same button that worked on the last machine you used - or stimulus-based learning, focusing on a distinctive feature, like a blinking light, and selecting that feature regardless of what physical action is needed to operate it.
Both strategies can succeed, but they have different strengths. Action-based approaches are fast and automatic when situations resemble past experience. Stimulus-based approaches are more flexible, allowing evaluation of options before committing to a movement. When it is unclear which strategy is better suited to the current situation, the brain needs some mechanism to arbitrate between them - to assign weight to each strategy based on which has been more reliable recently.
"The key distinction is whether learning should be tied to a motor action or the identity of the stimulus," said senior author Alireza Soltani, associate professor of psychological and brain sciences at Dartmouth. "Action-based learning involves considering the specific motor movements that can lead to a reward, while stimulus-based learning can be more flexible because it allows you to evaluate and select a desired stimulus without immediately considering the actions needed to get there."
Computational models and patients with amygdala damage
The team developed computational models based on reinforcement learning to track how the brain assigns weight to each learning strategy under uncertainty. They applied these models to behavioral data from patients with amygdala damage and healthy controls performing decision-making tasks.
In healthy participants, the amygdala appeared to pivot between the two systems early in a task, before enough information had accumulated to favor one strategy. As evidence built, it shifted weight toward the more reliable model. When the amygdala was damaged, this arbitration became more random. The brain also showed a consistent default toward action-based learning from the outset, losing the flexibility to explore whether stimulus-based learning might be more appropriate. Behavior became more rigid overall.
"A healthy amygdala promotes exploration between alternative models and, as a result, can make you choose something you wouldn't otherwise choose, and you can learn from that," Soltani said. The findings also help explain a pattern in older research: amygdala damage sometimes impairs stimulus learning while in other cases appearing to improve it - results that confused researchers for years. The arbitration model offers a coherent explanation: the impact depends on which strategy the context demands and how reliably the damaged amygdala can assign weight between the two.
What this means for phobias and anxiety
The reframing has practical implications for how clinicians think about fear-based disorders. Someone afraid of spiders tends to fixate on the spider itself - a pattern driven by stimulus-based learning, where the fear is tied to a specific object and becomes rigid over time. Standard cognitive approaches often try to reframe the spider as less threatening, which also operates through the stimulus channel.
The Dartmouth research suggests that redirecting attention to action-based exploration - not toward the spider but toward a series of deliberate, repeatable behaviors in the spider's presence - might engage the amygdala's arbitration function more effectively, shifting the brain toward a more flexible learning mode.
"In this way, the amygdala, one of the brain's arbitrators, can favor action-based learning, which is a more reliable predictor of the outcome," Soltani said. "This promotes exploration and flexibility, helping to overcome fear even when the stimulus initially carries strong negative associations."
Next steps
The team is now analyzing neural recordings in the prefrontal cortex to understand how neurons there interact with the amygdala during arbitration. In collaboration with UCLA researchers, they are also conducting experiments with rats to examine specific neural pathways connecting the amygdala and prefrontal cortex. Those experiments should help determine whether the computational arbitration mechanism the models predict corresponds to identifiable patterns of neural activity.