The problem
Choosing the direction of a faint visual pattern and judging whether that choice is correct are related tasks, but success at one does not guarantee success at the other. Which evidence-processing steps and errors are shared, and which belong specifically to confidence? This self-evaluation is called metacognition.
Researchers call the perceptual choice a type-1 decision and the confidence judgment a type-2 decision. Both can require weighing how strong the evidence is. Computational noise is variability in that evaluation, represented in a fitted model; its magnitude does not by itself identify the biological cause.[1] [2]
Scope
Shared and distinct computations underlying human perceptual decisions and confidence, including the sources of computational noise in each.
What remains open
Confidence uses information about the decision, but it can also be affected by errors in evaluating that information. Similar error levels in carefully matched tasks do not show that the answer and confidence come from one identical process.[1]
Zheng and colleagues matched the need for fine-grained evidence evaluation in confidence judgments and choices influenced by expectations or rewards. Their modeling found comparable computational-noise magnitudes. Equal magnitudes can arise from different causes: the result does not settle whether the computations or neural mechanisms are identical, or establish the same relation across other kinds of self-evaluation.[1]
Why it matters
Feeling unsure can prompt another look; feeling sure can encourage you to proceed. Understanding when confidence tracks accuracy helps explain how we decide to check or change an answer, while allowing that a confident answer can still be wrong.[2]
Confidence helps determine whether to seek more information or commit to a choice. An error in reading the image and an error in evaluating the choice could both lead to misplaced certainty, but they imply different explanations of that behavior. Separating them connects perceptual evidence to how people monitor and adjust their decisions; it does not make confidence an infallible test of correctness.[1] [2]
Definitions
2 terms
metacognition
Evaluating one’s own cognitive performance, such as judging how likely a perceptual answer is to be correct. Confidence can guide checking without guaranteeing accuracy.[2]
computational noise
Variability in the computations used to evaluate evidence, as represented by a model. A fitted amount of this variability is not a direct measurement of a particular neural cause.[1]
Sources
- Type-1 and type-2 decisions feature computational noise of similar magnitude
Notes on Type-1 and type-2 decisions feature computational noise of similar magnitude
Matched evidence-evaluation demands yield similar modeled noise; the authors explicitly leave whether noise sources are identical open.
- Towards an integrative neuroscience of metacognition
Notes on Towards an integrative neuroscience of metacognition
Reviews convergent behavioral, computational and neural evidence for confidence and self-evaluation; candidate building blocks do not settle every mechanism.