Metacognitive

Metacognitive intelligence is the capacity to observe, evaluate, and regulate one's own thinking. It is knowing that you don't know. It is catching yourself in a bias before it shapes your conclusion. It is the scientist who distrusts her own result precisely because it confirms what she hoped to find.

AI systems can be made to express uncertainty — calibrated confidence scores, ensemble disagreement metrics, abstention thresholds. But these are engineered properties, not self-awareness. A model that outputs "I'm not sure" has not experienced doubt. It has produced tokens that pattern-match to expressions of uncertainty in its training data.

The difference matters because genuine metacognition changes behavior from the inside. When a human recognizes that she is reasoning poorly, she can step back, reframe, seek outside perspective. A model cannot step back from itself. It has no self to step back from.

Articles in this tier explore the oversight of one's own cognitive processes — the capacity that allows humans to improve their own thinking in real time.