Hindsight Bias
Hindsight bias is a cognitive bias in which knowledge of an outcome leads to overestimation of one's prior ability to have predicted it. In AI evaluation, it refers to the systematic tendency of models to exploit future information present in training or test data, causing them to perform better in retrospective scenarios than in genuine prospective ones.
Core Facts
Timeline (last 90 days)
A study published on arXiv quantified the phenomenon of hindsight bias in clinical temporal reasoning.
When models are exposed to the complete timeline, they exhibit stable, hindsight-sensitive answer shifts, tending to provide answers consistent with known outcomes but not reasonable at the decision point.
The hindsight bias was formally established by psychologist Baruch Fischhoff in 1975 through a series of experiments.
Hindsight bias can be subdivided into three subtypes: memory distortion, the illusion of inevitability, and the illusion of foreseeability.
Hindsight bias involves a memory reconstruction mechanism: upon learning the outcome, the brain automatically integrates the outcome information into the existing knowledge structure, making cues consistent with the outcome more salient.
All Facts (5)
Hindsight bias involves a memory reconstruction mechanism: upon learning the outcome, the brain automatically integrates the outcome information into the existing knowledge structure, making cues consistent with the outcome more salient.
65%UnverifiedA study published on arXiv quantified the phenomenon of hindsight bias in clinical temporal reasoning.
50%UnverifiedWhen models are exposed to the complete timeline, they exhibit stable, hindsight-sensitive answer shifts, tending to provide answers consistent with known outcomes but not reasonable at the decision point.
50%UnverifiedThe hindsight bias was formally established by psychologist Baruch Fischhoff in 1975 through a series of experiments.
50%UnverifiedHindsight bias can be subdivided into three subtypes: memory distortion, the illusion of inevitability, and the illusion of foreseeability.
50%