Why Pedagogy Fundamentals Matter More Than Ever — Lessons from Generative AI and Education

Article 7 of 9 in a series on pedagogy fundamentals in the AI age.

For two years I have been arguing with educators on both sides of an exhausting line.

On one side: schools that bought a tutor license and now their job is "to facilitate."

On the other: teachers running oral exams in candle-lit rooms because the AI ruined writing.

Both groups are exhausted. Both have been sold a story that takes their craft away.

There is no AI-proof classroom. There is only an AI-thoughtful classroom.

Not "AI-free." Not "AI-saturated." AI-thoughtful. The choice was never ban-vs-allow. The choice has always been substitute-vs-augment.

The difference between a classroom where AI erodes thinking and a classroom where AI deepens it is not the technology. It is the pedagogy.

The book that names this with most clarity is Generative AI and Education by B. Mairéad Pratschke (Springer, 2024). A small book that fits in a coat pocket. Pratschke refuses both panic and hype. She points the camera back at the only variable any of us actually control: learning design.

The pedagogy hasn't changed. The cost of getting it wrong has.

"From language education to library science, from coding to composition, GAI is poised to change practice in every field."

The scope is real. The speed is real. The disruption is real.

What is not real — and what gets quietly assumed in every "the rules have changed" conversation — is that the pedagogy has changed. It hasn't.

The science of how a human being learns something durably did not change in November 2022. ChatGPT did not invent a new kind of brain. The mechanisms — generation, attention, memory, motivation, sense-making, metacognitive monitoring — have been steady for forty years.

What changed is the consequence of getting the design wrong.

She goes harder, citing Bates:

"AI developers have been largely unaware that learning is developmental and constructed, and instead have imposed an old and less appropriate method of teaching based on behaviourism and an objectivist epistemology."

The people building your AI tools are mostly building them on a model of learning that learning scientists abandoned in the 1970s. When you use those tools without your own pedagogy in the room, you are not "embracing the future." You are quietly importing the past with a chrome finish.

What Pratschke actually says

On the substitution trap.

"Without this deliberate focus on critical analysis—on sense-making—AI can substitute for critical reasoning and thereby reduce student agency to take control of their own learning."

The artifact still looks fine. The output is better than what the student would have produced alone. That is exactly the problem. The visible outcome went up while the invisible learning went down. We graded the visible outcome.

On the assessment arms race.

"Schools in the USA banned the use of ChatGPT and Australian universities reverted to pen-and-paper exams. The latter changed tack just over a week later, deciding with the benefit of some deliberation that ChatGPT could be used but only if cited as a source."

A week. The "AI-proof exam" lasted a week.

"Detectors proved unreliable, even discriminatory, producing high rates of false positives and tending to target content written by those whose first language is not English."

The arms race didn't just fail. It actively harmed the students it was meant to protect — disproportionately the multilingual, the non-native English speakers, the students already at the margins.

"Assessment is the canary in the coalmine that institutional leaders ignore at their peril."

The cheating panic was a symptom. The disease is that for decades we have been assessing things AI can do — recall, summary, formulaic argument structure, surface fluency. AI didn't break assessment. AI revealed that a lot of assessment was already broken.

On bans.

"Students reporting being afraid to use GAI in their work—not because of a fear of the technology itself but because of a fear of being accused of cheating."

A ban-first culture doesn't produce thoughtful AI use. It produces fearful AI use, hidden AI use. It disproportionately harms the most anxious students — the rule-followers, the over-checkers.

"Policing students is not only a poor approach in terms of trust, it is also financially unsustainable, and the conversation about assessment is not really about assessment at all."

On hallucination as gift.

"Students working with GAI need to develop their ability to make sense of what GAI generates and to locate it within wider contexts, viewing GAI as a critical opponent in disguise as a sycophant assistant."

A critical opponent in disguise as a sycophant assistant. Confident wrongness is the dominant rhetorical mode of the internet, of social media, of half of corporate PR. Hallucination is a free, infinite, perfectly calibrated supply of confident wrongness.

It is a teaching gift. We just have to use it.

On generative learning.

"Generative learning is a sense-making activity. It involves making sense of learning materials… by actively organising and integrating it with one's existing knowledge."

Sense-making is active. It happens in the gap between a stimulus and a stored mental model. AI can generate stimuli. It cannot do the gap-crossing for the student. The only way you can see whether the student did it is to make the process visible.

On the new role of the teacher.

"The educator is the disciplinary expert in this constructivist, connectivist, generative and social learning community, who can help students interpret output and discern between hallucinations and useful information, and hone their critical and analytical skills."

You are no longer the source of content. You are the mediator of content. A mediator who hides their own practice from the people they are mediating for is not actually mediating. They are performing.

What you do on Monday

1. Substitution Audit. Walk through any AI-allowing task as a student would. Where is the cognitive work? Name the thinking move. If a student handed this step to AI, would the artifact still look acceptable? If yes, that step is at risk. Redesign it so the student's thinking has to leave a trace — a reasoning log, a margin annotation, a five-minute oral defense. Re-prompt the AI's role: don't ask AI to do the step; ask AI to interrogate the student's attempt.

"You are a thinking partner, not a writer. The student will share a draft. Your only job is to ask three questions that pressure-test the weakest claim, surface a counter-example, and probe the connection between evidence and conclusion. You may not rewrite. You may not summarize. Only question."

2. Process-First Assessment. Stop grading only the artifact.

• Capture the trail. A "thinking log" — drafts, dead ends, prompts used, decisions made.
• Defend in dialogue. A five-minute live oral. What was hard? What did you change your mind about? Where did AI help and where did it mislead you?
• Show the iteration. First-attempt and final artifacts both submitted.
• Grade the learning, not just the polish. At least a third of the grade on process evidence.

An "AI-proof" prompt is impossible. An AI-transparent process is more pedagogically honest than what most of us were doing before AI existed.

3. Class AI Charter. In the first week of any AI-involving unit, run a 60-minute co-design session.

• Map the task. Where would AI help? Where would it hurt? Where would it cheat us out of learning we actually want?
• Define four zones. No AI (the live thinking moves). AI as critic. AI as assistant. AI as full collaborator.
• Write it down. A one-page charter, in their words. Posted in the room.
• Revise it openly. Every four weeks. The kids see that the rules of AI use are learned, not handed down.

The ban surrenders the conversation. The charter holds it. Students don't need you to police them. They need you to teach them how to police themselves.

4. Hallucination Hunt. Build verification into every research task as a graded element:

• Generate. AI produces the artifact in the open. Save the chat.
• Mark suspicion. Color-code every claim: definitely true, probably true, unverifiable, likely hallucinated.
• Verify. Find or fail to find a corroborating primary source.
• Annotate the failure modes. Why did the AI hallucinate this?
• Revise the artifact. Resubmit with hallucinations corrected.

The hallucination hunt teaches in one assignment what twenty lectures on epistemic humility cannot.

5. Visible Sense-Making. For every AI-enabled task, externalize the sense-making.

• Pre-prompt write. "What I think I already know. What I expect the AI to say. What I am unsure about."
• Generate. Save the conversation.
• Reaction write. "Here is what surprised me. Here is what confirmed my prior thinking. Here is one specific claim I am skeptical about."
• Integrate. Connect this material to the student's prior unit, prior life, prior experience.
• Submit all four artifacts together. The bundle is the assignment.

You are no longer grading the AI's output dressed in the student's name. You are grading the gap-crossing.

6. Equity-Visible Lesson. Treat AI access as visible curriculum, not invisible privilege.

• Audit. Could a student with no AI access do this well? If no, you have an equity problem in the design.
• Provide the floor. If AI is needed, the school provides access during class time.
• Teach the literacy in class. Prompt-craft, hallucination-spotting, substitution-audit — taught during lessons, not as homework.
• Make use visible. Students declare what tools they used and how. Normalize disclosure.
• Talk about the divide explicitly. A 20-minute lesson once a year on what unequal AI access means for their generation.

A student with a subscription and no AI literacy is barely better off than one with no subscription. The literacy is the great equalizer if we choose to build it.

7. Teach the Workflow. Make your own AI use visible, narrated, deliberate.

• Live-prompt in class, occasionally. Project the AI. Type the prompt. React out loud.
• Show your revisions. AI's first version and your revised version, side by side. Narrate what you changed and why.
• Disclose your tools. Small footer on handouts. Drafted with [tool], revised by [you].
• Run a "teacher AI fail" segment. Once a month: Here is a moment this week the AI got me wrong, here is what I noticed, here is what I changed.

If your students never see you use AI, they will assume good adults don't. They will hide their own use. The substitution loop will close in the dark.

The slogan

You are no longer the deliverer of content. You are the mediator of content. Not lecturer. Not gatekeeper. Not policeman. Mediator.

Pratschke near the end of her book:

"Far from limiting students' learning, when used in the context of generative learning's sense-making modes, GAI can be an 'engine of engagement.'"

An engine of engagement. That is what this technology can be — if the pedagogy holds. If the pedagogy slips, it becomes the opposite. An engine of erosion.

The technology is the same in both classrooms. The pedagogy is what differs.

Be the kind of teacher who, when your students grow up and look back, they say: that was the one who taught me how to think with the machine, instead of letting the machine think for me.

That is the only AI-proof outcome there ever was.

If this is landing, the EDodo flagship — AI-Powered Learning Design — is the cohort version of all this. Eight weeks of project-based building, peer review, real artifacts.

If you've been running substitution audits, charters, visible sense-making in your own classroom — and you speak fluent AI on top of it — please consider teaching with us.

Source: Pratschke, B. M. (2024). Generative AI and Education: Digital Pedagogies, Teaching Innovation and Learning Design. Springer. All quotes verbatim.