Why do some people get trapped in cycles of worry and avoidance, even when things are otherwise going well? For individuals with anxiety or depression, this persistence of negative expectation is not simply a matter of feeling negative events more deeply. Research in computational psychiatry suggests that it reflects a difference in how the brain learns from experience. The anxious brain may literally update its expectations from negative outcomes more rapidly than from positive ones, creating a world that feels persistently unpredictable and threatening.
Learning As A Computational Process
Everyday decisions, whether to speak up in a meeting, send a text, or cross the street, involve a process called reinforcement learning. This is the process by which humans and other animals adapt behavior based on feedback from the environment. Each feedback or outcome we receive from the environment slightly shifts how we value our choices. A learning rate measures how quickly a person updates their expectations after receiving new information. A high learning rate means that each outcome, good or bad, has a large impact on future decisions, while a lower rate reflects more gradual, averaged learning. Anxiety and depression, long described in emotional terms, can also be viewed through this computational lens.
Over-Learning From Punishment
In a study published in Nature Human Behavior, Jessica Aylward and colleagues (2019) asked healthy volunteers and people with mood and anxiety disorders to play a computerized “bandit” game. On each trial, participants chose between four slot machines whose rewards and punishments (i.e., happy or fearful faces) changed gradually over time. Their goal was to learn which options were currently most advantageous.
Using computational modelling, they found that anxious and depressed participants did not feel punishments more strongly, but they learned from them faster. After a negative outcome, their choices shifted, as if each bad outcome carried too much statistical weight.
A large meta-analysis by Pike and Robinson (2022) confirmed this pattern across 27 studies. Pooling data from more than 3,000 participants, they reported that people with mood and anxiety disorders have a faster learning rate from punishment and slightly slower learning rate from rewards. The findings suggest that symptoms such as avoidance or pessimism may reflect how the brain processes feedback, giving negative feedback a louder voice. Over time, this imbalance can distort behavior: each criticism or failure triggers rapid relearning, while rewards have less influence. The result is a feedback loop that reinforces expectations of loss.
Mechanisms Of Avoidance
A related study by Mkrtchian and colleagues (2017) examined not just how people learn from outcomes but how they decide whether to act at all. Participants completed a task that required either pressing a button or withholding a response while they were occasionally under threat of an unpredictable electric shock. The design allowed researchers to measure avoidance bias, the intuitive tendency to hold back when anticipating something unpleasant.
For individuals with mood and anxiety disorders, stress amplified this built-in reflex to inhibit behavior in the face of possible punishment, even when acting could have led to reward. In anxiety, systems that normally help us pause before danger appear tuned too conservatively, biasing the individual toward inaction and limiting opportunities to learn that feared situations are, in fact, safe.
Beyond Simple Learning Theories
The findings from reinforcement learning studies fit into a broader discussion written by Collins and Cockburn (2020). They advise that describing our decision systems as simple opposites risks missing the true complexity. Human learning is not governed only by positive and negative outcomes but by many interacting processes: memory, attention, emotion, and context all shape how feedback is used.
From this perspective, anxiety-related changes in punishment learning or avoidance are shifts in how multiple subsystems, such as valuation, prediction, inhibition, coordinate when uncertainty and threat are high.
Implications For Symptoms And Treatment
Together, these studies show that individuals with mood and anxiety disorders learn differently from experience in ways that can maintain their symptoms. The combination of rapid updating after punishment, slower integration of rewards, and a bias toward inhibition may help explain why anxious individuals struggle to benefit from reassurance, find it difficult to recover from failure, and often avoid uncertain situations.
Quantitative models such as these do more than describe behavior, they suggest new targets for therapy. Pike and Robinson (2022) suggest that treatment should not focus only on helping people be less distressed by negative outcomes, but on changing how they respond to them. Interventions might aim to slow the rapid behavioral updating that follows punishment, encouraging patients to pause before changing course so that a single setback carries less weight. In practical terms, this means helping individuals learn to trust that one painful experience does not determine what comes next.
References
Aylward, J., Hales, C. A., Robinson, O. J., & Robinson, O. J. (2019). Altered learning under uncertainty in unmedicated mood and anxiety disorders. Nature Human Behaviour, 3(10), 1116–1123.
Collins, A. G. E., & Cockburn, J. (2020). Beyond dichotomies in reinforcement learning. Nature Reviews Neuroscience, 21(10), 576–586.
Mkrtchian, A., Aylward, J., Dayan, P., Roiser, J. P., & Robinson, O. J. (2017). Modeling avoidance in mood and anxiety disorders using reinforcement learning. Biological Psychiatry, 82(7), 532–539.
Pike, A. C., & Robinson, O. J. (2022). Reinforcement learning in patients with mood and anxiety disorders: A meta-analysis. JAMA Psychiatry, 79(8), 765–775.