Metacognitive Information Theory.

Area of Science:

• Cognitive Neuroscience
• Decision Sciences
• Psychology

Background:

• Metacognition, the ability to assess one's own cognitive states, is crucial for regulating thought and action.
• Existing measures like meta-d' and M-ratio for metacognitive sensitivity and efficiency rely on reverse-engineering accuracy from confidence judgments.
• Brain regions involved in metacognition are linked to the quality and regulation of cognitive processes.

Purpose of the Study:

• To introduce a straightforward measure of metacognitive sensitivity, termed meta-I (meta-Information).
• To propose normalized versions of meta-I for quantifying metacognitive efficiency.
• To present an alternative to existing measures that may not fully capture sensitivity and efficiency.

Main Methods:

• Meta-I quantifies the mutual information between choice accuracy and confidence reports.
• Two normalized meta-I versions were developed to assess efficiency across different regimes.
• The proposed measures were illustrated using data from a perceptual decision-making task and a simulated metacognitive observer.

Main Results:

• Meta-I provides a direct assessment of metacognitive sensitivity.
• The proposed efficiency measures offer nuanced quantification.
• Unlike other metrics, meta-I-based measures improve with increased correctly assessed confidence bins.

Conclusions:

• Meta-I and its normalized variants offer a robust and intuitive framework for measuring metacognition.
• These novel measures enhance the assessment of how confidence reflects actual performance.
• The findings contribute to a deeper understanding of metacognitive processes in decision-making.