# Learning Withdrawal Drill · 学习撤回练习

> Use when AI has made learning feel easy but retention, calibration, or independent judgment is
> weakening. The drill decides what to keep inside the stop-line and how much friction to reinsert.

## 1. Capacity Under Review

| Field | Fill |
|---|---|
| Skill / capacity |  |
| Current AI use |  |
| Why it matters downstream |  |
| Risk if it atrophies |  |
| Human owner |  |

## 2. Outsourceability Test

| Question | Yes / No | Evidence |
|---|---|---|
| If this capacity weakens, do downstream judgments get worse? |  |  |
| Does the person need it to detect AI mistakes? |  |  |
| Is the skill constitutive of professional identity or taste? |  |  |
| Can the output be machine-verified without human understanding? |  |  |

Decision: `handoff / spar with AI / keep inside stop-line / park and self-learn`

## 3. Friction Setting

| Level | Use when | Rule |
|---|---|---|
| 0 · all-hand | Fully delegable and verifiable | AI answers first. |
| 1 · light | Low-stakes, learner has weak base | Learner predicts first, then checks. |
| 2 · band | Capacity must be retained | Learner drafts / reasons first; AI verifies second. |
| 3 · max | High-stakes expert capacity | AI only critiques after independent work. |

Chosen level:

## 4. Drill Plan

| Week | Human-first task | AI role | Reflection artifact | Review signal |
|---|---|---|---|---|
| 1 |  |  |  |  |
| 2 |  |  |  |  |
| 3 |  |  |  |  |
| 4 |  |  |  |  |

## 5. Retention Signal

Re-test without AI on:

Pass condition: