A person can solve a difficult problem quickly, explain the reasoning clearly and still be poor at recognizing the answer that deserves a second look. Another person may solve fewer problems but have an unusually accurate sense of which answers are solid and which are guesses.

Those are not contradictory profiles. Psychologists distinguish performance on a task from metacognition, the ability to monitor and evaluate one’s own thinking. The two are related, but the research does not support the tidy idea that greater intelligence automatically brings perfect intellectual self-knowledge.

The distinction is more useful than the familiar slogans about intelligent people always doubting themselves or less capable people always being certain. Neither is a reliable rule.

Intelligence and self-knowledge overlap without becoming the same thing

Intelligence is already a difficult word because it can refer to several things. In psychometrics, it usually means performance across tests of reasoning, vocabulary, memory, spatial processing and related abilities. Scores on these tasks tend to correlate, producing the broad statistical factor called general intelligence.

I wrote earlier about that positive pattern across mental tests. It is one of the most replicated findings in intelligence research, but it does not mean every mental ability is interchangeable. A broad tendency can coexist with meaningful differences between people and between tasks.

Metacognition asks a different question. After someone makes a judgment, can that person tell when it is likely to be right? Researchers often study this by asking participants to give both an answer and a confidence rating. The interesting part is not confidence alone. It is whether confidence rises for correct answers and falls for incorrect ones.

The second judgment hidden inside every answer

Imagine two people who each answer eight questions correctly out of ten. One is highly confident about all ten answers. The other gives high confidence to the eight correct answers and hesitates over the two mistakes. Their task scores are identical, but the second person has shown better sensitivity to the quality of their own decisions.

Researchers call that metacognitive sensitivity. It differs from a person’s general confidence level, sometimes called metacognitive bias. Someone can be modest about every answer yet still discriminate well between stronger and weaker judgments. Someone else can sound certain without that certainty tracking accuracy.

This is why verbal caution is a poor stand-in for intelligence. Saying “I might be wrong” can reflect genuine error monitoring, social politeness, anxiety, habit or strategic modesty. Sounding decisive can reflect knowledge, overconfidence, communication style or a situation in which the evidence really is clear. The words alone do not reveal the underlying ability.

A meta-analysis found only a weak connection

The strongest broad result comes from a 2018 meta-analysis by Kazuhiro Ohtani and Tetsuya Hisasaka in the journal Metacognition and Learning. Their analysis covered 149 samples drawn from 118 articles and examined the relationships among metacognition, intelligence and academic performance.

Overall, metacognition was only weakly correlated with intelligence. The size of the relationship also changed depending on how researchers measured metacognition. The authors found that metacognition predicted academic performance even after controlling statistically for intelligence, suggesting that monitoring and regulating thought may contribute something beyond raw performance on ability tests.

That result does not establish two sealed systems in the mind. It says the overlap is limited and measurement matters. Academic metacognition can include planning, checking comprehension and deciding when to change a learning strategy, while laboratory experiments may ask whether confidence tracks correct perception. Those are related ideas, but they are not identical tasks.

Knowing your limits may depend on the domain

A person who judges memory accurately may not be equally good at judging visual perception, factual knowledge or social reasoning. In a 2018 review in Personality Neuroscience, Marion Rouault, Andrew McWilliams, Micah Allen and Stephen Fleming found mixed evidence for a single metacognitive ability shared across domains.

Some studies found that metacognitive efficiency carried across perceptual and memory tasks when performance was carefully matched. Others did not. The review also noted that perceptual metacognitive efficiency was unrelated to fluid-intelligence measures in several datasets, including a large dataset of roughly 1,000 people.

The emerging picture is not that intelligent people lack self-awareness. It is that self-evaluation is partly local. Years of feedback can make someone well calibrated in one field while leaving them poorly calibrated in another. Expertise may improve both the first answer and the ability to recognize a doubtful one, but that calibration does not automatically travel everywhere.

Even measuring metacognition is difficult

There is a technical complication beneath all of this. A measure of how well confidence tracks accuracy can be distorted by the person’s underlying task performance. If one task is much easier for one participant than another, comparing their confidence patterns becomes statistically awkward.

Matthias Guggenmos tested widely used measures against simulations and the Confidence Database, a collection containing 145 studies and data from 6,912 participants. His 2021 paper in Neuroscience of Consciousness found that none of the current measures was fully independent of first-order task performance. Reliability also depended heavily on how many trials an experiment contained.

That does not make the field meaningless. It makes simple verdicts risky. When a study claims that one group has better insight into its own thinking, researchers have to ask whether the difference reflects metacognition, task difficulty, measurement noise or some combination.

The Dunning-Kruger slogan is too simple

This topic often gets reduced to the Dunning-Kruger effect, usually summarized online as “stupid people think they are smart.” That sentence is not an adequate account of the research, and it turns a measurement problem into an insult.

The original work concerned how poor performers estimated their performance on specific tasks. Later researchers have debated how much of the familiar pattern reflects metacognitive limits and how much can arise from regression toward the mean, noisy self-estimates and the mathematical boundaries imposed by dividing people into performance groups.

In a 2019 paper in the Journal of Experimental Psychology: General, Robert McIntosh and colleagues used trial-by-trial judgments in movement and spatial-memory tasks to separate task skill from self-assessment more carefully. Their results supported some relationship between skill and insight while challenging the usual global interpretation. A separate 2022 analysis by Jan Magnus and Anatoly Peresetsky showed that a statistical model could reproduce much of the classic pattern without assuming a psychological deficit.

The safest conclusion is narrower than the meme. People differ in both cognitive performance and the accuracy with which they monitor that performance. Those differences can be related, but one does not perfectly reveal the other.

An intelligent answer and an accurate sense of confidence are two achievements. Sometimes they arrive together. Sometimes they do not.