−17%
High-school students given unrestricted GPT-4 scored 48% higher in practice but 17% lower on the exam they then took without it. A hint-only tutor version largely mitigated the harm.
Bastani 2025 · About 1,000 students at one school in Turkey.
How students use AI matters more than whether they do
Recent studies show AI both helping and harming learning. What made the difference was the design of the teaching around it.
319
In a survey of 319 knowledge workers, more confidence in AI went with less critical thinking, and more self-confidence with more.
Lee 2025 · Self-reported survey of adults, not students.
46%
In PISA 2022 creative thinking, 46% of Korean students were top performers, yet their creative self-efficacy sat below the OECD average.
OECD 2024 · Rank is a range (2nd to 4th of 64).
The Compass: three pillars, twelve skills
The OECD and European Commission AI Literacy Framework separates using AI from understanding and managing it, and seven in ten employers in the World Economic Forum's survey call analytical thinking essential. CIT synthesised that research and those frameworks into twelve skills. At the centre, curiosity, resilience and agency are coached in every lesson and never graded. OECD & European Commission 2026 · World Economic Forum 2025
Think
Break problems down, see the structure, test with evidence.
-
T1
Computational Thinking
컴퓨팅 사고
Computational thinking is breaking a problem into steps, keeping only what matters, writing rules that solve it, and checking that those rules actually work.
At CIT The procedure goes on paper before the editor opens.
-
T2
Systems Thinking
시스템 사고
Systems thinking is seeing causes and effects as connected loops, and finding the delays, the stakeholders and the leverage points where a small change makes a large difference.
At CIT Every project starts with a drawing of causes and effects joined by arrows.
-
T3
Critical & Scientific Thinking
비판·과학적 사고
Critical and scientific thinking is testing claims, including an AI's answers, against evidence and controlled experiments instead of accepting them as given.
At CIT A prediction is written before the AI answers, and every difference is underlined.
-
T4
Data & Probabilistic Thinking
데이터·확률적 사고
Data and probabilistic thinking is reasoning from data under uncertainty, and reading what a model metric such as accuracy does and does not tell you.
At CIT Students build a model on data they collected and count its hits and misses by hand.
Create
Start from people and build all the way to something used.
-
C1
Empathy & Human-Centred Design
공감·인간중심 디자인
Empathy and human-centred design is starting from real people's needs and testing what you build with those same people.
At CIT Students interview the people who will use the thing before building it; the AI only transcribes.
-
C2
Creative Ideation
창의적 발상
Creative ideation is diverging by hand first, extending with AI, then converging by human judgment, so the result stays original.
At CIT Ten ideas go down by hand before the AI is opened.
-
C3
Builder Agency
빌더 주도성
Builder agency is shipping, from nothing to finished, an artifact that real people actually use.
At CIT Every unit ends with one thing that actually works.
-
C4
Integrative Thinking
융합적 사고
Integrative thinking is connecting disciplines and, when two of them pull against each other, finding an answer that keeps what matters in both.
At CIT Projects attach AI to the field the student already cares about.
With AI
Decide what to delegate, verify it, own the result.
-
A1
Specification & Promptcraft
명세·프롬프트 사고
Specification and promptcraft is stating the goal, context, constraints, format and acceptance criteria whenever you hand work to a machine or a person.
At CIT A five-box spec is filled in before any code is handed to an AI.
-
A2
Delegation & Verification
위임·검증
Delegation and verification is splitting work between human and AI through Plan, Delegate, Verify, Review, bounding every AI task and checking everything it returns.
At CIT The task is split into human, AI and shared parts in a table before work starts.
-
A3
Trust Calibration
신뢰 보정
Trust calibration is knowing, with reasons, when to trust an AI and when to trust yourself.
At CIT Students write how sure they are before checking, then weigh it against the AI's answer.
-
A4
Ethical & Futures Judgment
윤리·미래 판단
Ethical and futures judgment is weighing who a creation is fair to, whom it could harm and what it leaves behind over time, and crediting the AI's part openly.
At CIT Every artifact is published with its AI-use log attached.
The structure: from first principles to courses
The Compass has four tiers. First-principles thinking and metacognition sit at the apex and govern every skill; below them come the three pillars and the twelve skills; at the base, each course is one domain combined with the two or three skills it trains most.
- First Principles Thinking First-principles thinking is reducing a problem to what is known for certain, then rebuilding from there.
- Metacognition Metacognition is knowing what you know, what the AI knows, and when to trust which.
0 Apex
First Principles ThinkingEvery lesson opens with one first-principles question.
MetacognitionEvery lesson closes with a teach-back or a calibration check.
1 Pillars
ThinkCreateWith AI
2 Compass skills
3 Courses and applications
Course finder
Select a course to light up the skills it trains on the pyramid and the radar. Select a skill to list the courses that train it.
AI + X lab
- Domain
- Any discipline the student chooses
- Lab themes
- Space engineering with AI · Vision robotics · AI and psychology (Mind Lab)
- First-principles question
- Which field's assumptions is AI about to break?
- Roadmap track
- Project
Space engineering, vision robotics and the Mind Lab run as themes of this lab.
AI + X lab theme
Space engineering with AI
- Domain
- Orbital mechanics, mission design
- First-principles question
- What must be true for a satellite to stay up?
- Roadmap track
- Elite / Project
Opens first as an AI + X lab theme.
AI + X lab theme
Vision robotics
- Domain
- RoboMaster S1, Tello
- First-principles question
- What does the robot actually see?
- Roadmap track
- Leader / Elite
AI + X lab theme
AI and psychology (Mind Lab)
- Domain
- Adolescent neuroscience, behaviour
- First-principles question
- What is a feeling made of?
- Roadmap track
- Elite
AI olympiad track
- Domain
- ML theory, Python, NumPy, PyTorch
- First-principles question
- What does a model learn when it learns?
- Skills trained
- Roadmap track
- Elite / Project
Apps for real users
- Domain
- Web, product
- First-principles question
- Who has this problem and why?
- Roadmap track
- Leader / Elite
The course-to-skill mapping is CIT's own draft and will be confirmed with our teachers. Curiosity, resilience and agency are the dispositions under the whole structure; unlike the skills, they are never graded.
The twelve skills in depth
For each skill: what it is, why it matters with AI, how CIT teaches it, and what proficiency looks like at each age.
T1Computational Thinking 컴퓨팅 사고
Definition Computational thinking is breaking a problem into steps, keeping only what matters, writing rules that solve it, and checking that those rules actually work. It starts with procedures that give the same answer to the same input (CT 1.0) and widens to rules learned from data, whose answers come with uncertainty (CT 2.0).
Why it matters with AI
The more code an AI writes for a student, the more the student's job becomes deciding what to build and whether the result is right. Models learn their rules from data, so the same question can get different answers; reading a result well depends on knowing that.
How CIT teaches it: Paper algorithm first
Before opening an editor, students write the procedure as a flowchart or pseudocode, and a classmate runs it by hand from the page alone. Older students solve the same problem twice, once with hand-written rules and once with a model trained on data, and compare where each one fails.
Proficient looks like
- Elementary Writes a procedure of about five steps in order and, when a classmate follows it, finds and fixes the step that went wrong.
- Middle school Splits a problem into functions and writes tests, boundary cases included, before writing the code.
- High school Compares a hand-written rule with a trained model, using failure cases to explain the strengths of each.
T2Systems Thinking 시스템 사고
Definition Systems thinking is seeing causes and effects as connected loops, and finding the delays, the stakeholders and the leverage points where a small change makes a large difference. Before building, students ask what an AI would make better for the people involved. If they cannot answer, they conclude that AI does not belong there (the AI+X bridge test).
Why it matters with AI
Adding an AI tool changes what people do, and what they do becomes the tool's next data. Without seeing that feedback, a well-meant tool can produce results nobody expected.
How CIT teaches it: Causal loops and a leverage memo
Once a topic is chosen, students draw the causal-loop diagram. Then they write a one-page leverage memo: which arrow an AI would sit on, and what would change, for whom, if it did.
Proficient looks like
- Elementary Links causes and effects with three or more arrows and says what else changes when one of them does.
- Middle school Marks one reinforcing and one balancing loop and explains where an effect arrives late.
- High school Tabulates the effect on each stakeholder and argues, with evidence, for the cases where leaving AI out is better.
T3Critical & Scientific Thinking 비판·과학적 사고
Definition Critical and scientific thinking is testing claims, including an AI's answers, against evidence and controlled experiments instead of accepting them as given. To say why a result changed, change one condition at a time. The same rule applies to code, to experiments and to comparing AI answers.
Why it matters with AI
AI states wrong answers as confidently as right ones. In a survey of 319 knowledge workers, more confidence in AI went with less critical thinking. The habit of checking does not arrive on its own.
How CIT teaches it: One-variable experiments
Students change one condition, run it again, record what changed and explain it in their own words. When an AI answers, they compare it with their written prediction and name at least one way to check it.
Proficient looks like
- Elementary Answers 'why do you think so?' with one reason and restates a prediction that turned out wrong.
- Middle school Designs a one-variable experiment and marks, with reasons, where an AI answer is wrong.
- High school Weighs the reliability of sources, looks for counter-evidence and writes the limits of their own claim.
T4Data & Probabilistic Thinking 데이터·확률적 사고
Definition Data and probabilistic thinking is reasoning from data under uncertainty, and reading what a model metric such as accuracy does and does not tell you. It moves from reading averages and charts, through samples, distributions and kinds of error, to choosing the model metric that fits the situation.
Why it matters with AI
AI answers come out of probabilities. A model that is 95% accurate can still be useless when the data are lopsided. Deciding whether to trust a model starts with reading its numbers.
How CIT teaches it: Filling a confusion matrix by hand
Students train a model on a small dataset they gathered, count right and wrong cases in a table, then argue which mistake is more dangerous and which metric should therefore be watched.
Proficient looks like
- Elementary Reads the largest and smallest values from a chart and gives one reason for them.
- Middle school Distinguishes an average from a distribution and shows, with an example, how a small sample makes a conclusion shaky.
- High school Computes accuracy, precision and recall and justifies which one the problem calls for.
C1Empathy & Human-Centred Design 공감·인간중심 디자인
Definition Empathy and human-centred design is starting from real people's needs and testing what you build with those same people. Telling an AI's simulated empathy apart from the real thing is an explicit learning target of this skill.
Why it matters with AI
Chatbots produce a caring tone with ease, but a caring tone is not understanding. A design that never met a real user tends to fit the plausible user an AI made up.
How CIT teaches it: Interviews and an empathy map
Students interview at least three future users and draw an empathy map. They set a chatbot's 'empathetic' answer to the same questions beside the real interview answers and look for what differs.
Proficient looks like
- Elementary Asks a friend or family member three questions and sums up the answers in their own words.
- Middle school Turns one user need from the interview notes into a problem statement and revises the prototype on user feedback.
- High school Records assumptions apart from observations and explains, with evidence, how simulated AI personas differed from real users.
C2Creative Ideation 창의적 발상
Definition Creative ideation is diverging by hand first, extending with AI, then converging by human judgment, so the result stays original. The order is the point: people diverge first, AI extends second, and people choose.
Why it matters with AI
Stories written with AI help were rated more creative one by one, yet more alike as a group. Starting from AI pulls everyone toward the same few ideas.
How CIT teaches it: Ten by hand first
Students write ten ideas by hand before opening the AI, compare them with the AI's list and mark the ones that are theirs alone. An originality audit checks how far the final idea sits from existing work and from the AI's suggestions.
Proficient looks like
- Elementary Produces five or more ideas on one theme in pictures or words and explains the one they chose.
- Middle school Keeps hand-made and AI-made ideas in separate lists and writes the criteria behind the final choice.
- High school Uses an originality audit to show how the final idea differs from existing work and AI suggestions.
C3Builder Agency 빌더 주도성
Definition Builder agency is shipping, from nothing to finished, an artifact that real people actually use. It grows from finishing a small piece, through a second version revised on feedback, to gathering real users.
Why it matters with AI
AI has lowered the cost of a first attempt. What now sets a student apart is finishing the thing and getting someone to use it.
How CIT teaches it: Build to finish, then a second version
Each unit ends with a working artifact and a second version revised on user feedback. Older students publish theirs, keep a record of real use and set the next improvement from that record.
Proficient looks like
- Elementary Finishes a small game or piece and shows it to someone else.
- Middle school Ships a second version revised on user feedback and records what changed and why.
- High school Publishes the artifact, gathers real users and sets the next goal from the usage record.
C4Integrative Thinking 융합적 사고
Definition Integrative thinking is connecting disciplines and, when two of them pull against each other, finding an answer that keeps what matters in both. It starts with borrowing a concept from one field for another and grows into resolving two conflicting criteria while stating what the compromise costs.
Why it matters with AI
AI hands over knowledge from many fields at once. Deciding which pieces to combine, and how, stays with the person who owns the problem.
How CIT teaches it: AI+X studios
Students attach AI to their own field (music, sport, biology, finance) and build a project. They find where the two fields' demands collide, accuracy against privacy for example, and write a resolution and its cost.
Proficient looks like
- Elementary Connects two subjects, science and art for example, in one piece and says what came from each.
- Middle school Applies code and data to a problem in their own field and explains which concept came from which field.
- High school Resolves a conflict between two criteria and states the cost of the compromise.
A1Specification & Promptcraft 명세·프롬프트 사고
Definition Specification and promptcraft is stating the goal, context, constraints, format and acceptance criteria whenever you hand work to a machine or a person. It begins with saying what, for whom and when it counts as done, and grows into turning acceptance criteria into tests and fixing a spec's ambiguities unprompted.
Why it matters with AI
A vague request to an AI gets a vague answer. Deciding up front what 'done' means works the same with an AI as with a human team.
How CIT teaches it: The five-box spec
Students fill five boxes (goal, context, constraints, format, acceptance criteria) before delegating to an AI, then hand the same spec to a classmate and revise it until both produce the same result.
Proficient looks like
- Elementary Says in one sentence each what is wanted, for whom, and what counts as finished.
- Middle school Writes a five-box spec and turns its acceptance criteria into checkable tests.
- High school Finds and fixes ambiguities in a spec, and refines it by comparing what an AI and a person produced from it.
A2Delegation & Verification 위임·검증
Definition Delegation and verification is splitting work between human and AI through Plan, Delegate, Verify, Review, bounding every AI task and checking everything it returns. Plan → Delegate → Verify → Review is the canonical sequence for this skill.
Why it matters with AI
In a field experiment with consultants, tasks inside AI's frontier were done in greater number, faster and better, while on a task outside it consultants were 19 percentage points less likely to be correct. Knowing where to delegate decides the result.
How CIT teaches it: The delegation table and AI-use log
Students break a task into pieces, assign each to human, AI or both, and write the check for every AI piece in advance. Afterwards they review the AI-use log for any check that was skipped.
Proficient looks like
- Elementary Says which parts an AI or an adult helped with and which they did themselves.
- Middle school Writes a delegation table and logs, for each AI output, how it was checked and whether it was kept, revised or rejected.
- High school Explains from their own project where the AI was and was not reliable, and adjusts the delegation plan for the next one.
A3Trust Calibration 신뢰 보정
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.
Proficient looks like
- 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.
A4Ethical & Futures Judgment 윤리·미래 판단
Definition Ethical and futures judgment is weighing who a creation is fair to, whom it could harm and what it leaves behind over time, and crediting the AI's part openly. It begins with crediting help, moves to auditing bias and privacy, and reaches weighing effects years ahead.
Why it matters with AI
Once AI is involved, authorship blurs easily. The IB does not ban AI tools, but it requires AI-generated material to be credited.
How CIT teaches it: Impact audit and a public AI-use log
Every project gets a bias and privacy checklist, and a log of which AI was used at which stage and how, published with the artifact.
Proficient looks like
- Elementary Names one way their work could make someone uncomfortable and credits the help they had.
- Middle school Completes a bias and privacy checklist and attaches the AI-use log to the artifact.
- High school Writes an impact assessment that looks years ahead and answers the strongest objection from someone with a different stake.
The CIT Loop: six stages from Empathize to Share
Every project passes through six stages, each with a required artifact and a defined role for the AI. The six habits repeated inside every lesson (think, ask, verify, create, improve, defend) are set out on the Evidence page.
Safe walk to school Middle-school group · 8 weeks · sample design
A sample design written to explain the CIT Loop, not a completed student project. It will be replaced by a real case once a pilot cohort finishes.
-
1 Empathize 공감
Hear what the people who will use it actually live with, before building anything.
- Artifact
- Interview notes, empathy map
- AI's role
- Transcription only
Example · Weeks 1–2 Interview five classmates and two parents about where the walk to school feels unsafe. The AI transcribes the recordings; students do the sorting and the interpreting.
-
2 Map 지도화
Draw the problem as loops of cause and effect and find where a change would matter most.
- Artifact
- Causal-loop diagram, leverage memo
- AI's role
- Critic
- Main skills
- T2 Systems ThinkingC4 Integrative Thinking
Example · Weeks 3 Link vehicle speed, signal wait times and pedestrian numbers by time of day. Once the diagram is drawn, ask the AI which loops are missing; students decide which criticism to accept.
-
3 Compute 계산
Pin down what to compute in a spec and sketch the algorithm by hand before any code.
- Artifact
- Spec, algorithm sketch, test plan
- AI's role
- Tutor; code assistant only after the spec
Example · Weeks 4–5 Build a risk rule per street segment from public traffic data and pedestrian counts the students made, then write the five-box spec and the test plan. Until then the AI is a hint-only tutor; it becomes a code assistant only once the spec is done.
-
4 Create 창조
Build a prototype that works and check that the idea is genuinely the team's own.
- Artifact
- Prototype, originality audit
- AI's role
- Co-creator, with the mode declared
- Main skills
- C2 Creative IdeationC3 Builder Agency
Example · Weeks 6 Build a map web app that marks the risky segments, with an originality audit comparing the hand-made idea list against the AI's suggestions.
-
5 Critique 비평
Try to break what was built and ask whom it could disadvantage.
- Artifact
- Tests, bias and privacy audit, AI-use log
- AI's role
- Red-teamer
Example · Weeks 7 Run the tests and audit privacy, including whether the app stores students' locations. Ask the AI to find situations in which the app would give wrong directions, then verify each case it returns.
AI in the room, thinking by the student
Five routines run in every lesson. In a high-school maths experiment, an AI that handed out answers lowered exam scores while a hint-only tutor largely mitigated the harm, and a separate trial found a tutor built on pedagogical best practice beat in-class active learning. The difference came from design, not from the tool. Bastani 2025 · Kestin 2025
Think-First Window 먼저 생각하기
A set time alone with the problem, approach on paper, before any AI is opened.
Tutor-Mode Prompts 튜터 모드 프롬프트
The AI students use is set up to give hints and questions, not answers.
Verify Gate 검증 게이트
No AI output moves to the next stage until a written check has passed.
Calibration Check 신뢰 보정 점검
Confidence is written down before the answer is checked, then compared with the result.
AI-Off Checkpoint AI 끄고 점검
Each unit ends with solving and explaining without AI.
What changes with age
The same twelve skills, taught at the depth each age can carry. Use this to find your child's starting point.
| Age | New weight on | What students do | How AI is used | Related programmes |
|---|---|---|---|---|
| ElementaryGrades 1 to 6 | Thinking in steps, asking and listening, ideas by hand first, finishing small pieces | Less using AI tools alone, more understanding how AI produces answers, through play and experiments. | AI Genius track Programs | |
| Middle schoolGrades 7 to 9 | Causal loops, one-variable experiments, the five-box spec, delegation tables and AI-use logs | AI in tutor mode giving hints; outputs pass a verify gate before use. | AI Leader track AI education | |
| High schoolGrades 10 to 12 | Reading model metrics, connecting a field to AI, oral examinations, impact assessments that look years ahead | AI as a colleague, with the delegation plan and use log published and an oral explanation given without it. | AI Elite track Portfolio International schools |
How the Compass meets admissions and competitions
Where each Compass skill touches a school framework, an admissions element or a competition. Each row reads: element → Compass skills → the artifact CIT students keep.
International (IB, AP)
| Element | Compass skills | CIT artifact |
|---|---|---|
| IB Approaches to Learning: thinking skills | One-variable experiment log, originality audit | |
| IB ATL: research skills (information and media literacy) | Source-reliability table, AI-use log | |
| IB ATL: self-management (reflection) | Confidence log, reflection log, second-version record | |
| IB ATL: communication and social skills | Interview notes, specification | |
| IB Theory of Knowledge | Knowledge questions drawn from logs of verifying AI answers | |
| IB Extended Essay and Internal Assessment ↗ | Research question, data analysis, citations that credit AI use | |
| AP CSP AI Create Project (from 2027-28) ↗ | An AI application for real users, its spec and its test record |
Korea (curriculum, competitions)
Korea runs 1,141 AI focus schools from March 2026 and plans 2,000 by 2028. 머니투데이 2026-03-09
| Element | Compass skills | CIT artifact |
|---|---|---|
| 2022 national curriculum: creative thinking competency ↗ | Hand-made idea list, originality audit | |
| Knowledge-information processing competency ↗ | Source comparison, data analysis, verification log | |
| Self-management competency ↗ | Confidence log, second-version record | |
| Collaborative communication competency ↗ | Interview notes, specification | |
| Community competency ↗ | Impact checklist, causal-loop diagram | |
| School inquiry and performance tasks (the basis of 세특 records) | Inquiry experience and records students can use in their own school work. Academy lessons themselves are not entered in the school record. | |
| KOAI (run by KITPA) → IOAI national team selection ↗ | AI-off checkpoints, model-metric problem logs | |
| Science gifted and science high schools: inquiry | Experiment design, inquiry report |
Results you can see: the Thinking Profile
A termly report that records the twelve skills from the student's own work and shows them on one radar. It is in preparation; the example below uses a fictional student.
Sample student (grade 8, fictional)
Values
| T1 Computational Thinking | 3 · Proficient |
|---|---|
| T2 Systems Thinking | 2 · Developing |
| T3 Critical & Scientific Thinking | 3 · Proficient |
| T4 Data & Probabilistic Thinking | 2 · Developing |
| C1 Empathy & Human-Centred Design | 2 · Developing |
| C2 Creative Ideation | 3 · Proficient |
| C3 Builder Agency | 4 · Extending |
| C4 Integrative Thinking | 2 · Developing |
| A1 Specification & Promptcraft | 3 · Proficient |
| A2 Delegation & Verification | 2 · Developing |
| A3 Trust Calibration | 2 · Developing |
| A4 Ethical & Futures Judgment | 2 · Developing |
- A radar of the twelve skills, compared with last term
- Links to the work behind each rating
- One observed line from the teacher
- Two goals for next term
- 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.
Evidence and sources
Every number on this page comes from the sources below. Claims that can change, such as policy and exam timelines, are re-checked every month. Checked as of 2026-11-08
The CIT Thinking Compass is CIT's framework, synthesised from the research and international frameworks above. The framework itself has not been proven; pilot data and rubrics will be published here as they accumulate.
Show the full bibliography (25)
Research on AI and learning
- 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. About 1,000 students at one school in Turkey.
- Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R. & Wilson, N. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. CHI '25. Among 319 knowledge workers, higher confidence in GenAI went with less critical thinking; higher self-confidence went with more. Self-reported survey of adults, not students.
- 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. 194 Harvard physics students; short-term outcome.
- Dell'Acqua, F., McFowland, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F. & Lakhani, K. R. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013. Inside AI's frontier, consultants completed 12.2% more tasks, 25.1% faster, with more than 40% higher quality; outside it they were 19 percentage points less likely to be correct. 758 BCG consultants; adult professional tasks.
- Doshi, A. R. & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. Stories written with AI ideas were rated more creative individually but were more similar to each other. Short-story task with adult writers.
- 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. 117 university students, one writing task.
- Kosmyna, N. et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv:2506.08872 (preprint). An early EEG signal that starting essays with ChatGPT went with lower engagement; not peer-reviewed. Preprint; 54 participants, 18 in the final session; essay writing only.
Skills surveys
- OECD (2024). PISA 2022 Results (Volume III): Creative Minds, Creative Schools. Korea country note. Korean 15-year-olds scored 38 of 60 in creative thinking (OECD average 33), with 46% top performers (OECD 27%), yet reported creative self-efficacy below the OECD average. Rank is a range (2nd to 4th of 64).
- World Economic Forum (2025). The Future of Jobs Report 2025, chapter 3: Skills outlook. Analytical thinking remains the top core skill; seven in ten employers consider it essential. Employer survey.
International frameworks
- OECD & European Commission (2026). Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education. Final version, launched 18 June 2026. AI literacy is more than tool use: four domains (Engage with AI, Create with AI, Manage AI, Shape AI) and 19 competences. A framework, not an evaluation of any programme.
- Miao, F. & Shiohira, K. (2024). AI Competency Framework for Students. UNESCO. Twelve competencies in four dimensions, each at three levels: Understand, Apply, Create. A framework.
- ISTE (2016). ISTE Standards for Students. Seven student standards, including Innovative Designer and Computational Thinker. A framework.
Admissions and policy (dated)
- College Board. AP Computer Science Principles course redesign. AP Central. From fall 2027 (the 2027-28 school year) an AI Create Project replaces the Create performance task; students build an AI application for real users. The May 2027 exam still uses the current Create task; the final course description is due in spring 2027.
- International Baccalaureate (2023). Statement from the IB about ChatGPT and artificial intelligence in assessment and education. The IB will not ban AI software; students must credit AI-generated material and not present it as their own. Check the school's own academic integrity policy too.
- 머니투데이 (2026-03-09). 교육부 AI 중점학교 1,141곳 선정 보도. Korea's Ministry of Education selected 1,141 AI focus schools from March 2026, planning 1,500 in 2027 and 2,000 by 2028. Press report; the Ministry's own release was not located.
- IOAI. Countries and territories (accredited organisations). South Korea is listed; its accredited organisation is the Korea Information Technology Promotion Agency (KITPA), which runs KOAI. Accreditation can change each year; re-check before each campaign.
- 교육부 (2022). 2022 개정 교육과정 총론 (교육부 고시 제2022-33호). Six core competencies: self-management, knowledge-information processing, creative thinking, aesthetic sensibility, collaborative communication and community. General curriculum framework.
Foundational works
- Wing, J. M. (2006). Computational thinking. Communications of the ACM, 49(3), 33–35. Computational thinking as a fundamental skill for everyone, not only computer scientists.
- Meadows, D. H. (2008). Thinking in Systems: A Primer. Chelsea Green. Stocks, flows, feedback loops, delays and leverage points.
- Papert, S. (1980). Mindstorms: Children, Computers, and Powerful Ideas. Basic Books. Children learn powerfully by building things they care about.
- Resnick, M. (2017). Lifelong Kindergarten: Cultivating Creativity through Projects, Passion, Peers, and Play. MIT Press. Projects, passion, peers and play as the conditions for creative learning.
- Martin, R. L. (2007). The Opposable Mind: How Successful Leaders Win Through Integrative Thinking. Harvard Business School Press. Integrative thinking: holding two opposing models and building a better third.
- Willingham, D. T. (2007). Critical Thinking: Why Is It So Hard to Teach? American Educator, Summer 2007. Critical thinking depends on domain knowledge and cannot be taught apart from content.
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. Fast intuitive judgment (System 1) versus slow deliberate reasoning (System 2).
- Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive-developmental inquiry. American Psychologist, 34(10), 906–911. Introduced metacognition: knowledge and monitoring of one's own thinking.
Frequently asked questions
Is AI allowed in IB work?
The IB does not ban AI software. AI-generated material must be credited, and presenting it as your own is academic misconduct. CIT students attach an AI-use log to everything they make, so crediting AI in school work becomes routine. Schools set their own rules too; check them as well. International Baccalaureate 2023
From what age do students use AI tools?
Most generative AI services set a minimum age, and minors often need a parent's consent; check each service's terms. At elementary level the Compass puts the weight on understanding how AI produces answers rather than on using the tools alone.
Why do first principles and metacognition sit at the top?
Because each of the twelve skills is a way of applying two deeper habits: rebuilding from what is known for certain, and knowing what you know versus what the AI knows. So every lesson opens with a first-principles question and closes with a teach-back or a calibration check. Flavell 1979
Does this replace coding lessons?
No. The Compass is not a separate subject; it is practised on top of coding, AI and AI+X project lessons. Research showing that critical thinking is hard to teach apart from content knowledge supports that design. Willingham 2007
Wouldn't it be better to keep AI out entirely?
The research points to how AI is used rather than whether. An AI that handed out answers lowered exam scores; one designed to give hints largely mitigated that harm. So CIT keeps AI in the room, with routines such as Think-First and the Verify Gate around it. Bastani 2025
How does this connect to competitions?
Korea's accredited organisation for the International Olympiad in AI is KITPA, which runs KOAI. Competition problems lean on computational and data thinking, and AI-off checkpoints build the skill used in the exam room. See the KOAI programme page for the competition track. KOAI programme IOAI. Countries and territories accredited organisations
What changes with the AP CSP redesign?
College Board has announced that from the 2027-28 school year an AI Create Project replaces the Create performance task: students build an AI application for real users. The May 2027 exam keeps the current task. Specs, delegation tables and user testing, all practised in the Compass, line up with that project. College Board. AP Computer Science Principles course redesign. AP Central.
Is the self-check a test?
No. It is a checklist of current thinking habits; it does not measure ability or predict admissions outcomes. Answers are scored in your browser and not stored. Students under 14 should answer with a parent.
How will we see results?
The most direct evidence is the work each project leaves behind: specs, experiment logs, AI-use logs and demos. A termly Thinking Profile that summarises them is in preparation.
Can my child study online?
Yes. Alongside classes in Apgujeong, students in Korea and abroad take lessons online. Format and time zone are agreed in the consultation. Online lessons
Is the Compass a proven theory?
It is CIT's framework, synthesised from the research and frameworks cited here. Its own effect has not been proven; pilot data and rubrics will be published on this page as they accumulate.
Start with your child's thinking habits today
Three ways in, from the lightest to a full conversation.
- The one-page summaryThe Compass and the admissions crosswalk on one A4 page, to print or save as PDF.
- The 3-minute self-checkTwelve questions on current thinking habits. Nothing is stored.
- Book a consultationWe match a programme to your child's grade, school and interests, in English or Korean.
Parent seminar: From Prompt to Proof
Dates for the quarterly parent seminar are announced on our KakaoTalk channel. Send the word 'seminar' there to hear about the next one.