You find yourself in front of an unknown coffee machine. You’ve encountered this type before, more or less — there’s a blinking button on the left. Do you press it because it was blinking, or due to its position on the left? When the reward comes, your brain faces a dilemma: it cannot discern which aspect truly caused the outcome. Thus, it processes both hypotheses simultaneously and patiently awaits to see which one proves to be more dependable.
Deep within this mechanism, nestled in the temporal lobe, a pair of almond-shaped neuron clusters are engaging in a function that neuroscientists have often overlooked. The amygdala — traditionally viewed as the brain’s alarm system, responsible for quickened pulses and perspiring palms — is actually conducting a far more nuanced task. A study featured in Nature Communications suggests it serves as a mediator among conflicting learning systems, guiding the brain toward whichever world model is genuinely effective.
“There’s nothing particularly primitive in the brain, even within this area,” states Alireza Soltani, associate professor of psychological and brain sciences at Dartmouth College and the leading author of the study. “While the amygdala has been labeled as an emotional fear mechanism,” he explains, “the latest findings indicate a more multifaceted role — one that becomes apparent only when uncertainty is prominent.”
Soltani and PhD candidate Jae Hyung Woo crafted their research around a deceptively straightforward inquiry: how does the brain determine between various learning methods? In navigating rewards and penalties, we do not depend on a singular tactic. Action-based learning follows your actions — press the left button, receive coffee. Stimulus-based learning observes your surroundings — the flashing light heralded success. Both processes operate concurrently and gather evidence. The challenge lies in which approach the brain favors at any particular instant.
“The crucial difference,” Soltani remarks, “is whether learning should relate to a motor action or the nature of the stimulus.” Stimulus-based learning offers greater adaptability — enabling evaluations of choices prior to committing to movement. Conversely, action-based learning is quicker, more instinctual, and closely tied to the body. Neither approach is inherently superior; their efficacy depends on the task at hand. What the brain requires, therefore, is a means to decide between them.
To uncover that means, the team evaluated three groups of monkeys undertaking a probabilistic learning task specifically designed to be genuinely uncertain. Rewards changed unexpectedly. The monkeys were unaware, at the commencement of each new set of trials, whether rewards were linked to the objects they selected or where they made their selections — a characteristic of numerous real-life learning environments, where rules remain unknown until inferred. Some monkeys had normal brains. Others had damage to the amygdala. A third group had suffered lesions in the ventral striatum, a part associated with reward-driven learning.
The findings were distinctly categorized. Monkeys with ventral striatum impairments exhibited predictable responses: reduced stimulus-based learning and a tendency toward actions. However, their arbitration method — the mechanism weighing one learning system against another — was largely unaltered. They could still adjust which model the brain relied upon; they merely received a diminished signal through one pathway.
Monkeys with amygdala damage displayed a telling difference. Their arbitration process didn’t just tilt — it flattened. The brain could no longer update effectively which learning system held more reliability. It consistently favored action-based learning from the beginning and did not adjust as new evidence emerged. Behavior became rigid, even when the stimulus-based system would have been more adaptive. “A healthy amygdala encourages exploration of alternative models,” Soltani notes, “which can lead you to choose options you wouldn’t typically select, allowing you to learn from that.”
This helps to resolve a longstanding conundrum troubling amygdala research for years. Previous studies presented a perplexing landscape: damage to the amygdala sometimes hindered learning, while at other times it enhanced it. A traditional single-strategy framework could not explain this inconsistency. The computational model proposed by Woo and Soltani offers insight: the outcome relies on which learning system the task engages and the extent of evidence accumulated before the amygdala’s arbitration function activates. Under certain circumstances, relying more heavily on the action-based system can actually be beneficial — especially when the stimulus system has already identified an incorrect answer and needs to be displaced. The normal role of the amygdala is to navigate that conflict. Without its guidance, the brain becomes entrenched in whatever initial pathway it has chosen.
“Traditionally, the amygdala has been examined from a fear-learning viewpoint, and that has been extended to reward learning,” comments Woo. “Our primary hypothesis was that it must possess additional roles due to its vast connections to the rest of the brain.” These connections extend to the orbitofrontal cortex, the prefrontal cortex, and the ventral striatum — regions engaged in planning, valuation, and decision-making. They