Definition Trust calibration is knowing, with reasons, when to trust an AI and when to trust yourself.
Metacognition sits at the apex, above all twelve skills; this skill takes one part of it, deciding how far to trust an AI.
Why it matters with AI
In a high-school maths experiment, students with unrestricted GPT-4 scored 48% higher on practice problems but 17% lower on the exam they then took without it. That gap opens when students do not decide for themselves how far to trust the AI.
How CIT teaches it: Confidence log and trust table
Before seeing whether an answer is right, students write how sure they are and compare it with the outcome. When the AI and the student disagree, they log which one they trusted and why, and use the record to work out where the AI can be trusted more.
What proficiency looks like at each age
- Elementary When an AI's or a friend's answer differs from theirs, asks why and names one way to check.
- Middle school Records confidence before answering and notes afterwards whether they were over- or under-confident.
- High school Documents, case by case, why an AI output was or was not trusted, and explains from that record where the AI can be relied on.
Four recorded levels
- 1 Emerging Does it when an adult walks through the steps.
- 2 Developing Does it alone on familiar tasks; needs support on new ones.
- 3 Proficient Shows the grade-band indicator on an unfamiliar task, unaided.
- 4 Extending Teaches it to others or carries it into a new field.
Where this skill is used in the CIT Loop
- 5. Critique Try to break what was built and ask whom it could disadvantage.
- 6. Share Show it to the people it is for and explain it in your own words, without AI.
Evidence
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26), e2422633122. Unrestricted GPT-4 raised practice grades 48% but lowered exam grades 17%; a hint-giving tutor version largely mitigated the harm.
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X. & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. British Journal of Educational Technology, 56(2), 489–530. ChatGPT support raised essay scores but not knowledge gain or transfer, and may encourage metacognitive laziness.
- Kestin, G., Miller, K., Klales, A., Milbourne, T. & Ponti, G. (2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15. An AI tutor built on the same pedagogical best practices produced median learning gains more than double those of an in-class active-learning session.
Courses that train this skill
No course is linked to this skill yet.
Related skills
A1 Specification & PromptcraftA2 Delegation & VerificationA4 Ethical & Futures Judgment
First Principles Thinking: Every lesson opens with one first-principles question. Metacognition: Every lesson closes with a teach-back or a calibration check.