In a London lab, volunteers were measurably calmer when they knew an electric shock was coming than when they only thought it might be. The difference showed up in their sweat glands and in the size of their pupils. Cortisol in their saliva confirmed that the experiment as a whole was stressful, producing a hormonal response and not only reported discomfort. The body’s alarm ran hottest when the odds of a painful shock, as each volunteer judged them, hovered near fifty percent, and it eased as they grew confident either way. Those odds came from a computational model that read each volunteer’s running estimate from how they played a laboratory game.
The finding cuts against a deep intuition that bad news is the thing that damages us: that dread tracks danger, and that the stress response is a sensible gauge of how severe a threat is. In this task, the stress response followed how open the outcome still was.
The experiment and what it measured
Archy de Berker and colleagues gave forty-five volunteers a virtual game of rocks and snakes, with mild electric shocks to the hand painful enough to count as a real stressor. On each trial one of two rocks appeared, the volunteer guessed whether a snake lay underneath, and then the answer was revealed; a snake meant a shock. The probability of a snake under each rock changed as the task progressed, and participants had to track it, updating their estimates across 320 trials.
The odds were never fixed. The probability mapping shifted without warning every 26 to 38 trials, sometimes 90/10, sometimes 70/30, sometimes a pure 50/50, and within each stretch the snakes turned up by chance, so each participant’s uncertainty rose and fell over the session. At some moments a participant had a confident read on which rock was likelier to hide a snake; at others the odds felt like a coin toss. Confidence and doubt were both present, and measurably different, within each participant across the session.
Skin conductance and pupil dilation tracked what happened to the body as the probabilities shifted. Both are signs of arousal that bypass any deliberate attempt at self-presentation, and both rose and fell with the uncertainty estimates. Subjective stress ratings did the same. When participants estimated their shock probability at around fifty percent, the measures peaked, and they fell as estimates moved toward zero or toward one hundred, whichever way the confidence pointed. Feeling sure of a shock and feeling sure of no shock sat equally low on the curve, and alarm ran highest at “might get shocked.” A surprising outcome gave the body an extra jolt, but uncertainty raised arousal on its own.
Saliva cortisol could only confirm that the session as a whole was stressful; the moment-to-moment link to uncertainty came from skin conductance and pupil size. The alarm also appeared to be doing useful work: participants whose stress rose most steeply with uncertainty predicted the snakes correctly more often.
The machine that measured the uncertainty
The researchers didn't just track what participants reported feeling. They used a hierarchical Bayesian computational model, built on the mathematics used to describe how brains update beliefs under uncertainty, to reconstruct each participant's running probability estimate from their trial-by-trial behavior. At each moment in the task, the model inferred the probability distribution the participant was working with, judged from what their behavior implied.
The predictor of stress was a formal estimate of subjective uncertainty extracted from their trial-by-trial choices. So de Berker and colleagues were measuring something more specific than "people who feel anxious report being stressed": the uncertainty the model read from each participant's choices drove both the stress participants reported and the responses of their bodies. And the two went together: people whose reported stress was most sensitive to uncertainty also showed the biggest uncertainty-driven swings in pupil size and skin conductance.What the brain is doing
In the rock-and-snake task every shock was the same, so the stress system had no difference in severity to track. It rose and fell with how unresolved the prediction was.
The anterior cingulate cortex has been one of the most studied regions in the neuroscience of decision-making and control for the past several decades, and also one of the most debated. Its precise function is contested. Conflict monitoring, error detection, attention allocation, performance evaluation: different researchers have emphasized different aspects, and none of these characterizations is obviously wrong. What they converge on is a region acutely sensitive to the gap between what was expected and what happened. Prediction error, in various formulations, is the ACC’s core business.
One influential account, the Predicted Response-Outcome model, has the ACC attempting to predict all potential outcomes and signalling surprise when events deviate from what was expected, positive or negative. In standard reinforcement learning frameworks, positive and negative prediction errors track good and bad outcomes; this model’s signals track unexpectedness instead, covering events that occurred but weren’t predicted and events that were predicted but didn’t occur. Another carries the idea into cognitive control, arguing that the dorsal ACC decides whether, where and how much attention and effort to invest by weighing the expected payoff of control against its cost.
When the brain has resolved a prediction, even a bad one, it has an answer. In computational terms, the calculation can stop. What’s coming can be braced for; the motor and endocrine systems can prepare. A certain bad outcome is still a problem, but the brain knows the state of the world and can respond accordingly. An uncertain outcome, good or bad, leaves the computation running.
The brain’s prediction machinery returns a probability distribution that won’t collapse, and it keeps returning it; the ACC is a likely candidate for where this happens, though the shock study itself did not scan brains. What a person feels during that stretch may be the texture of a system computing without arriving at resolution.
The degree to which the probability is unresolved is a stress driver in its own right, separate from the thing you might be dreading: every snake meant the same calibrated shock, and stress still rose and fell with the odds.
Why the brain was built this way
Randolph Nesse’s smoke detector principle offers a plausible logic, though no one can watch that selection happen. When the cost of a false alarm is low and the cost of a miss is catastrophic, a well-calibrated system will produce many false alarms. Anxiety, Nesse argues, is a system calibrated this way, to minimize the probability of a fatal miss rather than subjective distress. The frequent minor suffering from false alarms is the cost of the design working as intended.
Apply this to uncertainty specifically: in environments where threat was physical and predators were real, the animal that became maximally vigilant at any evidence of possible danger, without waiting for it to be confirmed, was the one more likely to survive. Waiting for visual confirmation before initiating the stress response was a bad strategy. The stress response is supposed to fire early, on incomplete information, and “something might get you” is the trigger.
Seen through that lens, the brain’s preference for a known bad outcome over an open one makes sense. Confirmed threats could be acted on. Uncertain threats required sustained surveillance, which was expensive but affordable as long as the uncertainty was bounded, and many ancestral threats were. The predator either got you or it didn’t; seasons ended; illness resolved one way or the other. The probability distributions reached zero or one, and the computation could stop. If this reading is right, the system is tuned for uncertainty that ends, and it runs into trouble when resolution is withheld indefinitely.
The specific way modern environments break this
You’ve had a biopsy and are waiting for results. The stakes are real and the stress response is running, but the endpoint is defined: the results come in and the probability collapses to zero or one. This is acutely stressful, yet it is the kind of experience the stress system seems built to manage, because its structure is familiar: uncertainty with a natural resolution.
Now consider precarious employment. A gig contract, a performance assessment with no defined criteria, an “at will” arrangement that depends on consumer ratings that can change week to week. There is no test result or review date on the calendar, no moment at which the probability resolves. The surveillance never stands down, because there is never an output that says: okay, done, you either have the thing or you don’t. Job insecurity has detrimental effects on employees’ health, job attitudes and attitudes toward their organization, a pattern that holds when pooling 133 studies.
The construct of intolerance of uncertainty, or IU, was developed to characterize this: the degree to which a person finds any unresolved probability aversive, independent of what the probability is about. High IU does not mean overestimating the likelihood of bad outcomes. It means finding it intolerable that the outcome is unresolved at all. For someone high in IU, a scan result with a ninety percent chance of being fine does not bring ninety percent relief. The remaining ten percent can feel almost as heavy as bad news itself, because it is still open.
IU, developed by Mark Freeston, Michel Dugas, and colleagues, is measured by the Intolerance of Uncertainty Scale: twenty-seven items in the original 1994 version, later distilled into a 12-item short form with a stable two-factor structure. The scale does not ask about content, such as specific feared outcomes or probability estimates for particular bad events. It asks about the relationship to not-knowing as a general state.
What the Intolerance of Uncertainty Scale measures
The IUS doesn't ask what you're worried about. It asks things like: "Unforeseen events upset me greatly." "My mind can't be relaxed if I don't know what will happen tomorrow." "I can't stand being undecided about my future." "When I am uncertain, I can't function very well." No item refers to specific feared content, whether cancer, financial ruin or a relationship failing. They ask only how you relate to the state of not-knowing.
The short form's two-factor structure reflects two ways that uncertainty-intolerance manifests. The prospective factor captures people who experience any uncertainty as an active threat, a source of anxiety in its own right. The inhibitory factor captures people who freeze: uncertainty doesn't just feel bad, it prevents action. Both factors predict anxiety and worry, and both cut across diagnoses. As a clinical instrument, the IUS reaches widely: high scores turn up across GAD, OCD, health anxiety, depression and other anxiety disorders. IU correlates, on average, almost as strongly with symptoms of depression and OCD as with GAD, the pattern you would expect if one underlying trait feeds several disorders. A system that treats unresolved probability as inherently threatening will locate threatening content wherever it turns its attention.Even after an established CBT that targets IU, approximately twenty to thirty percent of GAD patients don’t achieve full remission, and in those patients IU remains elevated. That is why Dugas and his colleagues argue for working on the intolerance more directly.
What the misfire implies about what helps
Cognitive restructuring, a core tool of standard CBT, asks whether the feared outcome is realistic: is the catastrophe you’re imagining likely? That’s a reasonable question, and it is aimed at the content, the specific probability estimate.
Reassurance-seeking tries to push the probability closer to zero. Thought suppression removes uncertain content from attention. Both try to make the probability feel more resolved, and both work briefly, until a new uncertainty arises and the loop restarts.
The person who repeatedly asks their doctor whether a symptom is serious is doing what their brain is asking for: trying to resolve the probability estimate. The problem is that reassurance-seeking can train the expectation that resolution is always available, so that each new uncertainty becomes a trigger for the seeking behavior because the last one was resolved by asking. The relief is real but brief: when patients were told after gastroscopy that nothing was seriously wrong, those high in health anxiety were worrying again within 24 hours, and still were a year later. High IU and reassurance-seeking travel together in GAD and health anxiety alike.
The alternative is IU-targeted CBT, developed since the 1990s primarily by the Dugas group, which exposes patients to the state of not-knowing rather than to feared outcomes. In behavioral experiments, the patient deliberately does not seek resolution and observes that the uncertainty is survivable. The question stays open, and nothing catastrophic follows. Over repeated trials, the relationship to unresolved probability shifts, and the aim becomes living with the open question.
In 28 adults with GAD who completed a ten-session CBT program, falling IU statistically accounted for more than half of the drop in worry over the course of treatment. That CBT never targeted IU directly; the researchers suspect that exposure to worry themes and cognitive restructuring built tolerance for unresolved probability along the way. Standard treatments may work partly by changing IU without ever naming it.
Pooled trials point the same way: treatment produces a large improvement in IU compared with control conditions, and the size of that change predicts how much symptoms improve.
A more specific therapeutic move is to separate uncertainties that can be resolved from those that structurally cannot. Someone waiting on a biopsy can confirm that the sample is being processed, when the results are due and who will call. They cannot confirm the result. Conflating the two, as if the result could be known if only they found the right person to ask, keeps the uncertainty system fully loaded across both. Separating them closes the questions that can be closed, the closest available approximation to the relief the brain is seeking.
The architecture of permanent unresolvability
The evolutionary argument was about bounded uncertainty, where the predator either got you or it didn’t. Much of modern life is built differently, and in some cases this isn’t an accident.
Social media platforms deliver social evaluation signals on a schedule that is, deliberately or incidentally, irregular. Likes arrive unpredictably, and comments may or may not materialize. A post goes out into a void and might resonate or might not, with no timeline for finding out, and there will always be another post after this one. In conditioning research, rewards that arrive intermittently and unpredictably build the habits that are hardest to extinguish. By the logic of the de Berker mechanism, that arrangement is what would keep the uncertainty computation running at low but continuous load. The probability never reaches zero or one, and the surveillance mechanism has no natural off switch. Among emerging adults, heavier social media use has been linked with higher anxiety. The association is correlational, but it fits what the lab work predicts for a source of social evaluation that never resolves.
Platform work adds its own layer to precarious employment: consumer ratings that fluctuate with factors outside the worker’s control, on top of income, status and whether there will be work next month, none of which resolves at a defined point.
News coverage often holds possible threats in the state the stress system handles worst: open. A story is still developing or an investigation still ongoing, and the threat stays real but unconfirmed. The path from news coverage to anxiety is more tangled and contested than for social media or job insecurity, but the framing itself fits the pattern.
The convergence across these environments isn’t, in most cases, the result of deliberate design. Nobody sat in a room and decided to keep human uncertainty permanently open. Platforms optimized for engagement, and engagement runs on reward learning, and people’s social media behavior follows the same reward-learning rules seen in the lab. Employment structures evolved to minimize employer risk and maximize flexibility, and news outlets competing for attention have every incentive to keep a story open. The result is a set of environments that resemble, in structure, the conditions that drove stress highest in the de Berker experiment.
The participants in that study eventually left the lab. After the 320th rock the session ended, the shocks stopped and there was nothing left to predict.
Gen AI Disclaimer
Some contents of this page were generated and/or edited with the help of a Generative AI.
Media
Key Sources and References
de Berker, A.O., Rutledge, R.B., Mathys, C., Marshall, L., Cross, G.F., Dolan, R.J., & Bestmann, S. (2016). Computations of uncertainty mediate acute stress responses in humans. Nature Communications, 7, 10996. https://doi.org/10.1038/ncomms10996
Alexander, W.H., & Brown, J.W. (2019). The role of the anterior cingulate cortex in prediction error and signaling surprise. Topics in Cognitive Science, 11(1), 119-135. https://doi.org/10.1111/tops.12307
Shenhav, A., Botvinick, M.M., & Cohen, J.D. (2013). The expected value of control: An integrative theory of anterior cingulate cortex function. Neuron, 79(2), 217-240. https://doi.org/10.1016/j.neuron.2013.07.007
Bomyea, J., Ramsawh, H., Ball, T.M., Taylor, C.T., Paulus, M.P., Lang, A.J., & Stein, M.B. (2015). Intolerance of uncertainty as a mediator of reductions in worry in a cognitive behavioral treatment program for generalized anxiety disorder. Journal of Anxiety Disorders, 33, 90-94. https://doi.org/10.1016/j.janxdis.2015.05.004
Hébert, E.A., & Dugas, M.J. (2019). Behavioral experiments for intolerance of uncertainty: Challenging the unknown in the treatment of generalized anxiety disorder. Cognitive and Behavioral Practice, 26(2), 421-436. https://doi.org/10.1016/j.cbpra.2018.07.007
Miller, M.L., & McGuire, J.F. (2023). Targeting intolerance of uncertainty in treatment: A meta-analysis of therapeutic effects, treatment moderators, and underlying mechanisms. Journal of Affective Disorders, 341, 283-295. https://doi.org/10.1016/j.jad.2023.08.132
Nesse, R.M. (2001). The smoke detector principle: Natural selection and the regulation of defensive responses. Annals of the New York Academy of Sciences, 935, 75-85. https://doi.org/10.1111/j.1749-6632.2001.tb03472.x
Sverke, M., Hellgren, J., & Näswall, K. (2002). No security: A meta-analysis and review of job insecurity and its consequences. Journal of Occupational Health Psychology, 7(3), 242-264. https://doi.org/10.1037/1076-8998.7.3.242
Vannucci, A., Flannery, K.M., & Ohannessian, C.M. (2017). Social media use and anxiety in emerging adults. Journal of Affective Disorders, 207, 163-166. https://doi.org/10.1016/j.jad.2016.08.040
Freeston, M.H., Rhéaume, J., Letarte, H., Dugas, M.J., & Ladouceur, R. (1994). Why do people worry? Personality and Individual Differences, 17(6), 791-802. https://doi.org/10.1016/0191-8869(94)90048-5
Buhr, K., & Dugas, M.J. (2002). The intolerance of uncertainty scale: Psychometric properties of the English version. Behaviour Research and Therapy, 40(8), 931-945. https://doi.org/10.1016/S0005-7967(01)00092-4
Carleton, R.N., Norton, P.J., & Asmundson, G.J.G. (2007). Fearing the unknown: A short version of the Intolerance of Uncertainty Scale. Journal of Anxiety Disorders, 21(1), 105-117. https://doi.org/10.1016/j.janxdis.2006.03.014
Gentes, E.L., & Ruscio, A.M. (2011). A meta-analysis of the relation of intolerance of uncertainty to symptoms of generalized anxiety disorder, major depressive disorder, and obsessive-compulsive disorder. Clinical Psychology Review, 31(6), 923-933. https://doi.org/10.1016/j.cpr.2011.05.001
Cheng, G.H.-L., & Chan, D.K.-S. (2008). Who suffers more from job insecurity? A meta-analytic review. Applied Psychology, 57(2), 272-303. https://doi.org/10.1111/j.1464-0597.2007.00312.x
Lucock, M.P., Morley, S., White, C., & Peake, M.D. (1997). Responses of consecutive patients to reassurance after gastroscopy: results of self administered questionnaire survey. BMJ, 315(7108), 572-575. https://doi.org/10.1136/bmj.315.7108.572
Lindström, B., Bellander, M., Schultner, D.T., Chang, A., Tobler, P.N., & Amodio, D.M. (2021). A computational reward learning account of social media engagement. Nature Communications, 12, 1311. https://doi.org/10.1038/s41467-020-19607-x
