Traditional interpretations framed the amygdala largely within the confines of emotional processing, anchored in fear responses and associated with reward learning in a reflexive, almost reactionary capacity. Jae Hyung Woo, lead author and doctoral candidate in psychological and brain sciences at Dartmouth, contends that considering the amygdala merely a repository for fear oversimplifies its functionality given its dense, widespread neural connections. These connections enable it to interact with key brain regions implicated in memory, sensory perception, and executive functions, suggesting a higher-order integrative role essential to adaptive behavior.

The research team delved deeply into how the amygdala contributes to learning under conditions of uncertainty, employing computational reinforcement learning models to simulate the brain’s internal arbitration processes. These models provided a nuanced perspective: when confronted with unfamiliar scenarios, the brain must choose between two learning strategies—action-based, which relies on replicating successful motor behaviors from past experiences, and stimulus-based, which hinges on identifying salient cues or features in the environment predictive of desired outcomes. Using the analogy of mastering an unfamiliar coffee machine, the dichotomy becomes clear—should one press a previously successful button (action-based), or focus on environmental signals like blinking lights indicative of readiness (stimulus-based)?

Alireza Soltani, senior author and associate professor of psychological and brain sciences, highlights the subtle but critical distinction between these modes. Action-based learning is tightly coupled with executing precise motor sequences, making it relatively rigid but effective when the sequence is well-known. In contrast, stimulus-based learning offers greater flexibility, allowing the individual to generalize across varying contexts by focusing on stimulus properties detached from specific motor actions. This flexibility is critical in novel or unpredictable environments, enabling exploration of alternative paths toward reward.

In their experiments, the researchers observed that a functioning amygdala dynamically arbitrates between these strategies, effectively ‘weighing’ the reliability of each model as more information becomes available. Initially, the amygdala samples both action and stimulus-based approaches, but as experience accumulates, it preferentially commits to whichever model demonstrates higher predictive accuracy. This adaptability facilitates efficient and context-sensitive decision-making, optimizing reward acquisition by exploiting the most dependable learning strategy at any given moment.

Crucially, lesions or damage to the amygdala disrupt this precise arbitration, leading to suboptimal learning and inflexible behavioral outcomes. Study participants with compromised amygdala function showed a marked bias toward defaulting on action-based learning from the beginning, failing to appropriately adjust their reliance on stimulus-based cues. This bias hampers exploration, fostering rigid patterns that can be disadvantageous in dynamic environments, aiding in explaining conflicting findings from previous research where amygdala damage either impaired or enhanced stimulus-driven learning depending on task demands.

This emerging framework positions the amygdala not as a vestigial remnant of primal fear but as a pivotal arbiter amidst complex neural computations, guiding the brain’s choices about how to approach uncertain learning problems. As Soltani summarizes, the amygdala “promotes exploration between alternative models,” allowing individuals to deviate from default behaviors and learn from novel experiences, ultimately finding the most reliable internal model to navigate reality effectively.

Beyond basic neuroscience, these insights carry profound implications for psychological treatment paradigms, especially in handling phobias and anxiety disorders. Patients often become locked in stimulus-driven avoidance behaviors—compulsively associating fear with specific stimuli such as spiders, generating rigid responses that are difficult to override. The new model suggests therapeutic strategies that encourage shifting attention away from the fear-provoking stimulus toward active behavioral exploration may be more effective. For example, gradually interacting with the object of fear through a series of controlled actions could engage the amygdala’s arbitration function, fostering flexible learning and diminishing maladaptive stimulus-locked fear responses.

Such an approach reframes exposure therapy by leveraging the amygdala’s ability to favor action-based learning pathways, which tend to produce more reliable outcome predictions and enable cognitive flexibility. This reframing could represent a significant advance in our understanding and treatment of anxiety, suggesting novel interventions that promote exploration and resilience even in the face of deep-seated fears.

Intriguingly, this refined understanding of the amygdala reflects its evolutionary trajectory, now appreciated as a brain structure rooted in survival yet adapted for increasingly abstract integrative processes. As one of the oldest brain regions, its foundational role in threat detection and immediate reaction has been augmented by flexible, deliberative functions afforded by expanded connections to cortical areas, including the prefrontal cortex. This expansion allows the amygdala to influence decision-making processes far beyond reflexive fear, integrating diverse streams of information to arbitrate between competing learning models dynamically.

Moving forward, the Dartmouth research team is pursuing further investigations into the interplay between the amygdala and prefrontal cortex during decision-making tasks, utilizing neural recordings and rodent models to dissect specific pathways that facilitate this arbitration. Their collaborative work with UCLA aims to unravel the intricate circuitry governing this process, shedding light on how higher-order brain regions communicate with ancient limbic structures to orchestrate adaptive behavior.

This research not only challenges long-standing dogma about the amygdala’s function but also opens new avenues for understanding cognitive flexibility, learning under uncertainty, and emotion regulation. It stands to transform both basic neuroscience and clinical practice by highlighting the amygdala’s sophisticated role as a neural arbitrator—a conductor integrating disparate learning systems to optimize human decision-making in a complex, often uncertain world.

Subject of Research: Not applicable
Article Title: Contribution of amygdala to dynamic model arbitration under uncertainty
News Publication Date: 28-Nov-2025
Web References: http://dx.doi.org/10.1038/s41467-025-66745-1
Keywords: Neuroscience; Behavioral neuroscience; Cognitive neuroscience; Adaptive systems; Brain structure; Amygdala; Cognition; Risk perception; Decision making

Exploration and exploitation represent fundamental cognitive strategies. Exploitation involves leveraging existing knowledge to maximize immediate rewards, while exploration is the risky venture into the unknown in search of potentially better outcomes. Historically, much emphasis has been placed on how the brain navigates this dilemma when pursuing rewards. Prior research has implicated prefrontal cortical circuits and subcortical regions in mediating exploratory behaviors driven by anticipated gains. Despite this, the mechanisms driving exploration when individuals attempt to avoid negative outcomes—such as losses or punishments—have remained vague, primarily due to the difficulty of disentangling neural signals in complex, emotionally charged scenarios.

The recent study by Reitich-Stolero, Aberg, Halperin, and colleagues breaks new ground by investigating exploration within mixed valence conditions—both gains and losses—while simultaneously capturing single-neuron activity in human participants. Utilizing a probabilistic learning task that interleaved reward and punishment trials, the research team meticulously recorded neural firing patterns from the amygdala and temporal cortex, regions traditionally linked to emotion processing and memory, respectively. Their approach enabled them to parse out how neurons adjust their signaling prior to decisions to explore in both positive and negative contexts.

A defining revelation of their findings is the identification of two distinct neural components driving exploration. First, a valence-independent rate signal was detected, reflecting a consistent neuronal coding of decision urgency or propensity to explore, regardless of whether the anticipated outcome was a gain or a loss. This suggests that some aspects of the brain’s exploratory machinery function in a generalizable manner. Second, more intriguing is the valence-dependent global noise signal observed preferentially in the amygdala during aversive, or loss-driven, trials. This heightened variability or “neural noise” correlates with an increased tendency to explore when trying to avoid negative outcomes, implicating a specialized amygdala mechanism modulating cautious or risk-sensitive behavior.

The amygdala’s engagement in exploration amid uncertainty links to its well-documented role in processing emotional salience and threat detection. Previous imaging studies have hinted at its involvement, but single-cell resolution provides unprecedented granularity on how neuronal firing patterns translate into adaptive or maladaptive decision strategies. Increased noise in neuronal activity could reflect an intrinsic computational mechanism enabling flexible switching between exploitation and exploration, especially under duress or when outcomes carry potential harm.

Behaviorally, participants exhibited an intriguing asymmetry: they explored more aggressively when confronted with the prospect of losses than gains, a phenomenon that aligns with psychological theories of loss aversion. The neural data align with this behavioral disposition by showing that the amygdala’s elevated noise levels drive this heightened exploratory behavior, suggesting that the emotional weight of potential losses intensifies the neural decision-making noise, resulting in more frequent shifts into exploration mode.

These discoveries carry profound implications for understanding psychiatric conditions characterized by dysregulated exploration-exploitation balance, particularly mood disorders such as anxiety and depression. Elevated amygdala activity is a hallmark in many affective disorders, and this study’s linkage of amygdala noise to increased exploration offers a mechanistic bridge between altered neural dynamics and maladaptive behaviors observed clinically. Patients exhibiting abnormal exploratory patterns, for example, excessive avoidance or risk-seeking in the face of threat, may be rooted in perturbed amygdala signaling.

Furthermore, the identification of two separable components—rate and noise—underlying exploratory decisions provides a conceptual framework that can inform computational models of decision-making. The rate code offers a stable signal guiding the initiation of exploration across contexts, while noise introduces probabilistic variability that may optimize learning and survival under uncertainty. Such dual mechanisms expand the traditional view, which has often emphasized deterministic or reward-driven factors alone.

This study also underscores the temporal cortex’s role in coupling with amygdala signals during exploration. Given the temporal lobes’ involvement in memory encoding, it’s plausible that exploration is not solely a current-state decision but is enriched by integrating past experiences to forecast outcomes. This coordination may enable the brain to weigh historical contingencies alongside emotional signals, thus refining the exploration-exploitation balance adaptively.

Technically, recording single neurons in awake human subjects engaged in complex decision tasks represents a formidable achievement, harnessing rare clinical opportunities presented during neurosurgical procedures. This invasive approach surpasses the resolution limits of functional neuroimaging and offers unparalleled insights into the microcircuit operations that govern cognition. The researchers’ use of probabilistic task design further allows disentanglement of neural patterns linked to reward versus punishment uncertainty, establishing a robust foundation for interpreting the functional significance of observed neural dynamics.

The implications of this research extend beyond understanding basic neuroscience. By illuminating how noise in the amygdala can amplify exploratory tendencies, particularly under threat, it sets the stage for novel interventions that modulate neural variability. Pharmacological agents, neuromodulation techniques, or behavioral therapies aimed at recalibrating amygdala activity hold promise for restoring balanced exploration in patients whose decision-making is skewed by affective disorders.

Moreover, these findings inspire a reevaluation of how computational models incorporate stochasticity and valence in decision processes. Traditional reinforcement learning models may need refinement to account for valence-dependent modulation of internal noise, moving closer to biological realism. This may ultimately lead to better predictive tools for human behavior and tailored approaches in artificial intelligence systems modeling human-like exploration.

In sum, this pioneering work delineates how the amygdala’s firing rate and intrinsic noise collaborate to modulate human exploratory behavior in environments laden with uncertainty and emotional valence. It advances a nuanced understanding of the neural computations underlying decisions to explore or exploit, especially in the context of loss avoidance. The convergence of neural, behavioral, and computational insights spotlights the amygdala as a critical node orchestrating adaptive decision strategies, while also hinting at mechanisms that may go awry in psychiatric pathologies. As exploration remains a cornerstone of learning and survival, unraveling its neural basis continues to be a paramount quest, now propelled forward by these compelling advances.

Subject of Research: Neural mechanisms of exploration and exploitation trade-off in human decision-making under mixed gain and loss conditions, with a focus on amygdala activity.

Article Title: Rate and noise in human amygdala drive increased exploration in aversive learning.

Article References:
Reitich-Stolero, T., Aberg, K.C., Halperin, D. et al. Rate and noise in human amygdala drive increased exploration in aversive learning. Nature (2025). https://doi.org/10.1038/s41586-025-09466-1

Image Credits: AI Generated