Formation and AI modes¶

Reading is indispensable. Reading alone is not formation.

Formation modes¶

R | Read / Encounter¶

Primary sources, rigorous textbooks, data, artifacts, works of art, historical documents, field observation.

Evidence: Annotated reading/field ledger

S | Study / Reconstruct¶

Derive, diagram, explain, retrieve, solve, compare, build concept maps, reproduce examples without aid.

Evidence: Closed-book reconstruction

P | Practice / Drill¶

Repeated exercises with feedback, spacing, interleaving, increasing difficulty, error taxonomy.

Evidence: Practice log + performance curve

M | Make / Investigate¶

Experiments, code, essays, designs, models, performances, repairs, interviews, research.

Evidence: Inspectible artifact + process record

D | Dialogue / Critique¶

Seminar, Socratic questioning, peer review, debate, listening, design critique, negotiation.

Evidence: Discussion/critique record

A | Apply / Serve¶

Use knowledge for a real person, institution, community, field site, or operational system.

Evidence: Stakeholder/field evidence

B | Break / Red-team¶

Construct counterexamples, edge cases, adversarial prompts, failure injections, alternative interpretations.

Evidence: Failure catalogue

C | Compare / Curate¶

Rank alternatives, study expert exemplars, defend criteria, revise taste.

Evidence: Blind ranking + curation statement

F | Forecast / Decide¶

Make probabilistic forecasts and consequential decisions before outcomes are known.

Evidence: Scored forecast / decision journal

T | Teach / Translate¶

Teach a novice, translate across disciplines/languages/media, create worked examples.

Evidence: Learner pre/post or audience evidence

V | Viva / Defend¶

Oral defense, perturbation, hostile but fair questioning, identify unresolved uncertainty.

Evidence: Examiner record

L | Longitudinal Retest¶

Return weeks/months later, often without AI or notes, and solve a changed problem.

Evidence: Delayed score / transfer artifact

AI modes¶

A0 | Human-only acquisition¶

No general-purpose AI during first reconstruction.

Purpose: Protect initial encoding, mental models, fluency, and error detection.

Pass condition: Fail if learner cannot explain from memory at required granularity.

A1 | AI as tutor / questioner¶

AI may ask questions, generate practice, give bounded hints, or simulate an examiner.

Purpose: Use AI to increase deliberate practice, not replace it.

Pass condition: Pass only if gains persist without AI.

A2 | AI as copilot¶

AI may draft alternatives, code, search, simulate, translate, or critique.

Purpose: Learn augmentation and comparative judgment.

Pass condition: Polish alone receives no credit.

A3 | AI as delegated agent¶

AI may execute multi-step work with tools under permissions.

Purpose: Learn orchestration, monitoring, and exception handling.

Pass condition: Learner must predict likely failure classes before run.

A4 | AI as adversary / red team¶

AI deliberately attacks claims, designs, incentives, security, and reasoning.

Purpose: Train robustness and epistemic humility.

Pass condition: Pass requires changing something important when attack succeeds.

A5 | Human defense / fallback¶

AI removed; learner defends decisions, key models, evidence, and uncertainty.

Purpose: Ensure the person owns the result rather than merely operating a system.

Pass condition: Pass only if learner can distinguish what they know from what the AI supplied.