The Confirmation Bias That Runs Your Reasoning, and How to Override It.

In 2005, Philip Tetlock published the results of a twenty-year tracking study that should have ended every confident forecast on television. He had followed 284 professional forecasters (economists, political scientists, intelligence analysts) as they made 82,361 predictions about world events.^[2] The experts barely beat chance. Worse, the more famous the forecaster, the worse the accuracy. The single strongest predictor of poor performance was not ideology or domain expertise. It was the tendency to fit new information into an existing framework rather than updating the framework itself.^[3]

That finding has aged well. In the two decades since, confirmation bias (the tendency to seek, interpret, and recall information in ways that confirm what one already believes) has been traced from a behavioural pattern observed by Francis Bacon in 1620 to a measurable neural signature visible on a magnetoencephalography scan.^[1][9] The bias is not a character flaw. It is a feature of how the brain processes evidence after a decision has been made. And the science now shows, with uncomfortable clarity, that intelligence does not protect against it.

Stanovich and West tested this directly across more than 1,400 participants: cognitive ability, measured by standardised tests, was statistically unrelated to myside bias, the tendency to evaluate arguments more favourably when they align with one's own position.^[5] The smartest people in the room are not the least biased. They are often the most skilled at constructing arguments for what they already believe.

The practical implications are not abstract. In Mendel's controlled experiment with 150 clinicians, psychiatrists assigned to a confirmatory information search strategy reached the wrong diagnosis 70% of the time.^[26] Those given a disconfirmatory strategy (actively seeking evidence against their initial impression) dropped to 27%. A balanced search strategy produced an intermediate 47% error rate. The same clinical presentation, the same patient symptoms, the same available data. The only variable was the direction in which the clinician searched. That matters because the clinician's intelligence was not the bottleneck. Their search strategy was.

The bias is not limited to medicine. Berthet's cross-occupational review found confirmation bias distorting professional judgment in management, finance, law, and criminal justice, with the worst consequences appearing in wrongful conviction cases.^[29] In financial markets, Park and colleagues tracked 502 investors and found that those exhibiting stronger confirmation bias traded more frequently, held higher return expectations, and achieved lower realised returns.^[30]

The brain encodes confirming and disconfirming evidence with equal fidelity. That sentence deserves emphasis because it contradicts the intuitive model most people carry: that biased people simply do not see the opposing evidence. Park and colleagues demonstrated in 2025, using MEG combined with psychophysics in 30 participants, that the posterior parietal cortex represents both types of evidence with comparable neural precision.^[15] The information arrives. It is encoded. The sensory machinery does its job.

The bias enters at a later stage. Rollwage's team at University College London published the clearest mechanistic account to date in Nature Communications, using magnetoencephalography combined with drift-diffusion modelling across three independent experiments with 76 participants.^[9] The critical finding: after an initial decision is made, the confidence signal associated with that decision selectively amplifies the evidence accumulation rate for confirming information, while driving the accumulation rate for disconfirming information to near-zero. The bias is not attentional. It operates during the weighting phase. The brain receives the counter-evidence; it simply stops counting it.

This is the computational equivalent of a factory quality-control system that logs every defect report but routes complaints about its own product line to a folder no one reads. The data exists. The routing ensures it never reaches the decision-maker.

The routing has an emotional architecture. In 2006, Westen and colleagues scanned 30 strongly committed political partisans with fMRI as they evaluated contradictory statements from their preferred presidential candidates.^[8] The result overturned the assumption that motivated reasoning is a reasoning problem. The dorsolateral prefrontal cortex (the region associated with deliberate analytical reasoning) was not engaged during the bias. Instead, the brain recruited ventromedial prefrontal cortex, anterior cingulate cortex, posterior cingulate, and insula: the emotion regulation circuit. When subjects reached a bias-consistent conclusion, the nucleus accumbens (the brain's reward centre) activated, producing a neurochemical reward for successful self-deception. Researchers call this motivated reasoning.^[8]

Westen's study used a highly partisan sample selected for extreme political commitment, and it predates modern pre-registration norms. But the core finding has proven durable. Kaplan and colleagues replicated it in 2016 with 40 participants, showing that greater belief resistance correlated with heightened default mode network activation.^[13] Lois and colleagues confirmed in 2024 that opposing ideological groups use identical neural machinery (vmPFC, ventral striatum, dorsal ACC) to reach contradictory conclusions from the same evidence.^[14]

The implication is structural: the brain does not reason its way to a biased conclusion. It feels its way there, and then rewards itself for arriving.