Over Precision Illusion

People act like they know more than they really do. They give too much weight to their own guesses and seem too sure.

Over-precision is the cognitive bias where agents assign excessively narrow confidence intervals to their beliefs, underestimating actual uncertainty. This leads to overconfident judgments and underestimation of external variability in belief updating.

A friend confidently tells you the drive to the airport will take exactly 25 minutes, with no allowance for traffic or delays. When the trip takes 45 minutes, they are genuinely surprised — they were so certain in their single estimate that they never considered variability.

An equity analyst constructs a DCF model and reports a target price of $47.32, implying a 90% confidence interval of ±$2 around that estimate. Historical backtesting of comparable models shows true valuation uncertainty warrants a ±$15 interval, but the analyst's over-precision illusion causes them to assign prior_precision far above empirical calibration, producing likelihood_signal_attenuation for incoming macro data that should widen the interval. When the stock prints at $31, the model's posterior_revision is minimal because the narrow confidence interval structurally downweights the disconfirming evidence, illustrating a severe asymmetric_update_magnitude failure.

People pick a single answer and then feel very sure about it. Small bits of new information do not change their mind much.

A belief_updating_architecture structural element constrains posterior variance by applying a high-confidence prior around the point estimate, producing asymmetric updating where likelihood inputs are downweighted. The architecture creates a weighting_asymmetry that limits evidence-driven variance expansion and biases confidence upward.

Ask for a range of possible answers instead of one exact guess. Practice revising your confidence when new facts appear.

Elicit and use calibrated probability intervals and update priors to reflect empirical error rates. Implement variance-expanding priors or probabilistic debiasing during belief aggregation.

Ignored contradicting evidence; Underestimated outcome variability; Poor calibration of predictions

An adversarial actor can exploit over-precision by feeding a target agent a single authoritative-seeming data point or framing early in a decision process, knowing the target's narrow prior will anchor tightly and resist subsequent correction. Disinformation campaigns can leverage this by presenting fabricated certainty artifacts (e.g., fake statistics with spuriously precise decimal values) to induce confident false beliefs that are structurally resistant to updating. In forecasting or negotiation contexts, an adversary can strategically withhold variance-revealing evidence, exploiting the target's existing over-precision to lock in suboptimal commitments.

Practitioners should adopt explicit interval forecasting protocols — always eliciting a low, central, and high estimate — and track calibration scores over time to empirically measure and correct overconfidence. Implementing variance-expanding priors (e.g., wider default confidence intervals derived from historical base-rate error) during belief aggregation structurally counteracts the narrow-prior weighting dynamic. Adversarial red-teaming and pre-mortem analysis, where agents are required to argue for why their confident estimate is wrong, directly challenge concentrated posterior distributions before consequential decisions are made.