Step-by-Step Teaching

Step 1: Establish the Core Question

The Central Problem:

There is a fundamental difference between two things:

🔑 Key Insight: These two goals are not the same, and they may actually conflict with each other.

---

Step 2: Understand How Professional Expertise Is Built

Traditional Learning Path:

Routine tedious work → Mistakes & corrections → Guided supervision → Judgment

• Expertise requires thousands of hours of hands-on engagement
• Junior professionals learn by doing, especially through repetitive, formative tasks
• Senior professionals develop judgment — the ability to make high-stakes decisions independently

🔑 Key Insight: Judgment cannot be shortcut easily. It is earned through struggle, not just exposure.

---

Step 3: Understand the Two Competing Theories About AI and Learning

Before the study, researchers identified two opposing possibilities:

Theory A — AI Accelerates Learning ✅

• Studies in radiology, legal education, and business problem-solving showed that AI embedded in training can help less-experienced workers improve independently
• AI acts as a scaffold that builds real skill

Theory B — AI Creates Dependency ❌

• Studies with software engineers, consultants, and students showed AI acts as a "temporary exoskeleton"
• Output improves while using AI, but gains disappear when AI is removed
• The worker never truly learned — they just borrowed the tool's capability

🔑 Key Insight: Both theories can be true — but for different people under different conditions.

---

Step 4: Learn the Experimental Design

Why this study is credible — it had 3 rare elements:

Setup:

• 133 patent lawyers across 11 law firms
• Randomly split: ⅔ got AI access (treatment), ⅓ did not (control)
• Tasks measured at 10 days and 90 days

Two types of tasks:

Task 1 — DRAFTING (AI allowed)
Measures: immediate productivity boost

Task 2 — REDLINING (NO AI allowed)
Measures: actual skill development

🔑 Key Insight: The redlining task is the critical test — it reveals whether AI usage built real skill or merely substituted for it.

---

Step 5: Understand the Results — Layer by Layer

Layer 1: AI Boosts Everyone's Drafting Performance

• At 10 days: +0.34 standard deviations (≈ 10 percentile points)
• At 90 days: +0.38 standard deviations (≈ 11 percentile points)

What improved:

Before AI:  [Poor] [Mediocre] [Average] [Good] [Excellent]
After AI:   [Poor↓] [Mediocre↑] [Average↑] [Good] [Excellent — unchanged]

🔑 Key Insight: AI lifted the bottom, but did not create excellence. It eliminated weak work, not average work.

---

Layer 2: The Redlining Task Reveals a Split

When AI was removed, the overall group still outperformed controls by +0.32 SD — but this hides a critical difference:

Junior lawyers' scores actually split:

More very low scores ↑
Fewer mediocre scores ↓
More good scores ↑
No change in excellent scores

🔑 Key Insight: Juniors became more polarized — some got better, some got worse. AI did not reliably build their foundational skill.

---

Step 6: Understand Why the Gap Exists

This is the most important conceptual step.

Why Seniors Benefited:

Seniors used AI as a "logic auditor" — not as a replacement for thinking, but as a challenger of their existing reasoning.

Senior behavior pattern:

• Skipped low-stakes sections
• Rebuilt claims from scratch
• Paired edits with explicit legal doctrines
• Asked: "Why am I making this change?"

Senior + AI = Existing expertise × Sharpened by AI challenge

Why Juniors Did Not Benefit:

Juniors lacked the foundational expertise needed to use AI critically.

Junior behavior pattern:

• Worked rigidly top-to-bottom
• Focused on surface-level synonym swapping
• Diagnosed problems but didn't fix them ("diagnose-without-execute")
• This pattern existed in both AI and non-AI junior groups

Junior + AI = AI executes the work → Junior never practices the skill

🔑 Key Insight: AI can only amplify existing expertise. Without a foundation, there is nothing to amplify — and the tool may actually replace the learning process itself.

---

Step 7: Synthesize the Core Framework

Here is the complete mental model from this study:

                    ┌─────────────────────────────────┐
                    │         AI ASSISTANCE            │
                    └────────────┬────────────────────┘
                                 │
               ┌─────────────────┴──────────────────┐
               ▼                                    ▼
    ┌─────────────────────┐             ┌─────────────────────┐
    │   SENIOR WORKER     │             │   JUNIOR WORKER     │
    │  (Has foundation)   │             │  (No foundation)    │
    └─────────┬───────────┘             └──────────┬──────────┘
              │                                    │
              ▼                                    ▼
    AI challenges and                   AI executes tasks
    sharpens judgment                   junior hasn't mastered
              │                                    │
              ▼                                    ▼
    ✅ Better work +                    ✅ Better work NOW
    ✅ Better judgment                  ❌ No skill gain
                                        ⚠️  Polarized outcomes

---

Step 8: Understand the Broader Implications

The Tension for Organizations and Educators:

Why This Matters Long-Term:

• Lawyers still need judgment in courtrooms, client meetings, and negotiations — places where AI cannot assist
• If juniors never build foundational expertise, there will be no future senior experts
• The pipeline of expertise could be quietly hollowed out even as short-term quality metrics improve

🔑 Key Insight: Organizations optimizing for output quality today using AI may be inadvertently defunding the expertise pipeline of tomorrow.

---

Step 9: Know the Study's Limitations (Critical Thinking)

🔑 Key Insight: Good research acknowledges its boundaries. These limitations point to future research questions, not failures.

---

Final Summary: The Three Things to Remember

1. AI improves output quality for everyone — but does not equally build skill in everyone.

2. Foundational expertise is the prerequisite for AI to enhance judgment. Without it, AI substitutes for learning rather than accelerating it.

3. Better work and better workers are different goals — and in the age of AI, they may require deliberately different strategies.