The Real Problem with AI in Higher Education
For two years, programme directors have been hearing the same message: "you need to integrate AI into your curriculum." And for two years, the most common response has been to bolt ChatGPT onto an existing case study and call it pedagogical innovation.
Let's be direct: having a chatbot write an essay, then asking students to "critique" the AI's output, is not pedagogical innovation. It's a proofreading exercise.
The problem isn't the absence of AI in your courses. The problem is that most proposed uses follow the same logic: the AI does the work, the student supervises. But pedagogy works the other way around.
Two Visions of AI in the Classroom
Vision A: AI as Assistant
The student asks, the AI answers. The AI summarizes an article. The AI generates an outline. The AI corrects a draft. The student becomes a supervisor of a tool.
This is useful. It's a professional skill. But it's not learning — it's delegation.
Vision B: AI as Adversary
The AI doesn't answer — it resists. It plays a CFO who won't share financials unless you ask the right question. A historical witness with their own version of events. An unhappy client who won't say why they're unhappy until you've earned their trust.
In this vision, AI doesn't do the work for the student. It creates the work.
What This Actually Looks Like
Take an HR management course. The topic: a salary dispute in a mid-sized company.
Traditional Approach
The professor distributes an 8-page case study. Everyone reads the same document. Students answer the same questions. The best groups produce well-structured answers that look remarkably like... what ChatGPT would produce in 30 seconds.
AI-as-Terrain Approach
The professor creates a scenario in 5 minutes. AI generates a complete company: org chart, financial data, characters. Each character has their own knowledge, blind spots, and things they won't say.
- The HR director says "everything is under control." She won't mention the collective resignation notice unless asked specifically about recent departures.
- The union representative is cooperative, but only trusts you after two or three honest exchanges.
- The CFO has the numbers, but doesn't see why he should share them with "junior consultants."
Students investigate. Each team chooses who to talk to, what questions to ask, in what order. Two hours later, two teams facing the same problem have different analyses — because they didn't talk to the same people or ask the same questions.
The deliverable isn't "the right answer." It's their answer, built on their investigation. ChatGPT can't produce it — because ChatGPT didn't conduct the investigation.
Why This Solves Three Problems at Once
1. Engagement
Students don't read a case — they live a situation. The difference is visible within 10 minutes: students who never participate in class find themselves strategizing interrogation approaches between sessions.
2. Cheating
When assessment is based on the process (who did you talk to? what hypotheses did you test? what information did you cross-reference?), cheating becomes meaningless. Copying another group's deliverable is useless — their investigation was different.
3. Assessing Real Competencies
You're no longer measuring the ability to structure an answer. You're measuring the ability to ask the right questions, manage a difficult relationship, synthesize contradictory information, and take a position when not everything is clear.
These are exactly the competencies employers say graduates lack.
"OK, But How Do I Actually Do This?"
Three paths, from simplest to most ambitious.
Path 1: One AI Character in Your Course (30 minutes)
Take a case you already use. Instead of distributing it as a PDF, turn it into a character.
Create a ChatGPT prompt: "You are Marie Dupont, HR Director of a 45-person company. You're dealing with a salary dispute. You know the following facts: [list]. You don't volunteer them — you only reveal them if the student asks the right question. You're cooperative but you protect management."
Students interrogate Marie instead of reading the case. That alone radically changes engagement.
Limitation: One character = one perspective. No triangulation possible.
Path 2: Multiple Characters, One Scenario (5 minutes with MEτiS)
Describe the problem. AI generates the company, the characters, their knowledge and their limits. Launch the session. Students investigate freely.
Each character has their own perspective. The HR director and the union rep don't tell the same story. It's up to the students to cross-reference, verify, and build their own understanding.
You monitor everything in real time: who talked to whom, what questions were asked, what clues were discovered. Zero free riders.
Path 3: Assessment Through Investigation
Go further: the final deliverable is graded on three pillars.
- Content — the quality of the analysis produced
- Investigation — the rigor of the inquiry (how many sources consulted, what questions asked, what clues cross-referenced)
- Integration — the ability to use what was discovered in the deliverable
A student who produces a beautiful analysis without talking to anyone scores low on investigation. A student who talked to everyone but can't synthesize scores low on integration. The score reflects the process as much as the product.
What This Is Not
To be clear about what this approach doesn't claim:
- It's not a replacement for lectures. It's a complement — the lecture provides the theoretical framework, the simulation provides the experience.
- It's not a tech gimmick. The AI is invisible to the student — they see characters, not algorithms.
- It's not limited to business. The same method works in literature, history, law, economics, and language learning.
- It's not time-consuming to prepare. 5 minutes of description, AI does the rest. Adjust if you want — or launch as is.
The Right Test
Ask yourself one question: does your use of AI make the student more active or more passive?
If AI summarizes, generates, corrects instead of the student — it makes them passive.
If AI creates a problem to solve, an interlocutor to convince, a situation to untangle — it makes them active.
The most useful AI in education isn't the one that helps students. It's the one that makes their life harder — intelligently.
Try It Now
MEτiS turns any course topic into an investigation universe. Create a scenario in 5 minutes, launch a session, see what happens.
Free plan. No credit card. Create my first universe →