Chapter 28 Exercises: Learning in the Age of AI

Chapter 28 Exercises: Learning in the Age of AI

These exercises are designed to help you develop a more intentional and effective relationship with AI tools in your learning. They apply to anyone who uses AI in their learning or professional work.

Exercise 1: The Fluency Illusion Test (30–45 minutes)

This exercise empirically tests whether AI explanations produce understanding or the illusion of understanding.

Step 1: Choose a topic you genuinely don't know well but want to learn. Something where you're starting from limited knowledge.

Step 2 (baseline): Before consulting any AI, write down everything you currently know about this topic. Don't look anything up. Just: what do you know, right now?

Step 3: Ask an AI assistant to explain the topic to you. Read the explanation once, carefully. Don't take notes.

Step 4 (immediate test): Close the AI chat. On a new document or page, write down everything you now know about the topic — from memory. Don't look at the AI's explanation.

Step 5: Compare your post-AI writing to the AI's explanation. What percentage of the explanation can you reproduce?

Step 6: Wait 24 hours. Without consulting any source, write down what you know about the topic again.

Step 7: Compare the 24-hour recall to the post-AI recall and to the original explanation. What survived? What disappeared?

Reflection: - Was there a significant gap between "it made sense while reading" and "what I could reproduce immediately afterward"? - Was there further significant decay at 24 hours? - What does this tell you about the difference between AI-generated comprehension and durable learning?

Most people find that AI explanations produce better immediate comprehension than studying without AI, but that retention at 24 hours is similar to or not dramatically better than other passive study methods. The mechanism that produces durable learning (retrieval practice, generation effect) is not activated by reading AI explanations.

Exercise 2: The Socratic Tutor Protocol (one topic, 30–60 minutes)

This exercise implements the Socratic tutoring approach described in the chapter.

Choose a topic you're currently learning or want to understand more deeply.

Step 1 — Your explanation (5–10 minutes): Without consulting any resource, write a 200–400 word explanation of the topic. Explain it as if to a knowledgeable friend. Include what you think you know, what you're not sure about, and where your understanding gets fuzzy.

Step 2 — The prompt: Open an AI assistant and write: "I'm going to give you my explanation of [topic]. Based on it, please ask me 4–6 probing questions that would distinguish someone who genuinely understands [topic] from someone who has just read a surface explanation. Here's my explanation: [paste your explanation]"

Step 3 — Answer the questions: Answer each question from memory, without consulting any resource. For each answer, assess: "I know this well / I'm somewhat sure / I'm mostly guessing / I have no idea."

Step 4 — Target the gaps: For the questions you answered poorly (mostly guessing or no idea), ask the AI for an explanation. After each explanation, ask the AI to ask you a follow-up question to check whether you understood.

Step 5 — Final self-test: After the full exchange, close everything. Write a new explanation of the topic from memory. What's different from your first explanation?

Reflection: - What gaps did the questions reveal that you wouldn't have noticed through passive reading? - How did producing your own explanation first change the quality of the AI interaction? - Is the post-exchange explanation meaningfully more complete than the pre-exchange one?

Exercise 3: Design Your AI Use Policy (45–60 minutes)

This exercise creates a personal, explicit policy for AI use across different learning contexts — replacing implicit, default behavior with deliberate choices.

Step 1: List the domains where you currently use AI: - Work tasks (writing, coding, research, analysis, etc.) - Academic learning - Skill development (language, programming, etc.) - Creative work - Decision-making

Step 2: For each domain, answer: - What is the primary purpose of this activity? (To produce output, to develop a skill, to understand something, to make a decision?) - Which parts of this activity are ones where AI is accelerating work without compromising development? - Which parts are ones where AI might be replacing cognitive work you should be doing?

Step 3: For each domain, draft a brief personal policy: - When do you use AI? - What does "generate first" look like in this context (what and how long before consulting AI)? - What will you never use AI for in this context?

Example policies: - Academic writing: "I outline my argument first, without AI. I draft key paragraphs without AI. I may use AI for grammar editing and to check for logical gaps after I've committed to an argument." - Coding: "I attempt any problem for at least 20 minutes before consulting AI. I use AI as a code reviewer ('what's wrong with this?') not as a generator. I understand every line of any AI-generated code before using it." - Research: "I use AI for literature orientation (get the lay of the land). I never cite anything AI has told me without primary source verification."

Step 4: Review your policies across domains. Are they consistent with your goals? Are they realistic to maintain?

Exercise 4: The Generation Comparison Experiment (one week)

This exercise directly tests the generation effect in your own learning.

The experiment: For one week, you'll alternate between two approaches to learning new material:

Method A (generate first): When you encounter a new concept, idea, or problem: 1. Spend 5–10 minutes attempting to understand, work through, or explain it yourself, without consulting any resource 2. Then consult your resource (textbook, AI, documentation, etc.) 3. Attempt to explain or apply it again without looking

Method B (resource first): For another set of comparable material: 1. Consult your resource immediately 2. Read and understand it 3. Then attempt to explain or apply it without looking

Testing: After three days, test your retention of both sets of material with a quick self-quiz.

Tracking: Note which method felt more effortful and which felt more effective (these may differ).

Reflection at end of week: - Which method produced better retention on the delayed test? - Which felt more difficult during the learning? - Which produced more specific identification of gaps? - What does this experiment suggest about how you should structure your learning going forward?

Exercise 5: AI Calibration — Spot the Error

This exercise develops the critical evaluation skill that makes AI safe to use.

Choose a domain you know moderately well — well enough to detect some errors.

Step 1: Ask an AI assistant to explain three to five things in this domain. Write the explanations down.

Step 2: Fact-check each explanation against reliable sources (textbooks, peer-reviewed sources, authoritative reference materials — not other AI systems).

For each explanation, note: - Was it fully correct? - Were there any errors, omissions, or subtle misleadings? - Would someone with less knowledge than you have detected the issues?

Step 3: In a domain you know less well, ask the AI to explain three to five things and have an expert (or reliable source) check the outputs.

Reflection: - How many errors did you find? Were they minor or significant? - Did the errors affect the overall impression of the explanation (i.e., did they make it seem right even though it was wrong)? - What does this experiment suggest about the appropriate level of trust to place in AI explanations?

Reflection Questions

After completing at least two exercises:

What is the most significant risk of AI assistance for your specific learning goals? (The fluency illusion? Skill atrophy? Epistemic dependency? The calibration problem?)
In which area of your learning do you currently use AI in a way that might be replacing cognitive work you should be doing yourself?
What would the Socratic tutoring approach add to your current learning practice? Is it realistic to implement it regularly?
How has your thinking about AI and learning changed as a result of this chapter? What's one specific practice you'll change?