AI in classrooms makes critical thinking more vital than ever, says education expert at Tartu

Helen Crompton, a professor of educational technology at Old Dominion University, argues that artificial intelligence entering classrooms makes critical thinking and healthy scepticism more important than ever before. She warns that making learning too easy undermines resilience, while also seeing AI as a powerful equaliser for developing nations. Crompton recently visited the University of Tartu to oppose a doctoral defence and deliver a public lecture on AI in higher education.

Artificial intelligence entering classrooms across the world is not a threat to education, but it demands a fundamental rethinking of how learning is assessed and what skills students truly need, according to Helen Crompton, professor of educational technology at Old Dominion University and executive director of the Research Institute for Digital Innovation in Learning.

Crompton, one of the most cited experts in her field who advises both the United Nations and the World Bank on ethical and evidence-based use of technology, visited the University of Tartu in Estonia recently to oppose a doctoral defence and deliver a public lecture on the state of AI in higher education.

AI as a global equaliser

One of Crompton's central arguments runs counter to the instinct of many Western academics: she views AI not as a tool that widens the digital divide, but as a rare opportunity for developing nations to catch up. «In sub-Saharan Africa, people are thrilled about AI,» she said. «It gives people possibilities that we in Western countries have always had, like a personal virtual tutor or a business analyst when starting a company.»

She pointed out that calls to slow AI development almost exclusively come from privileged positions, from countries that can afford to opt out. «Those calls come purely from a position of privilege, taking away from developing countries the chance to catch up with us.»

The end of the essay

On the question of assessment, Crompton is direct: traditional written essays have largely lost their purpose in a world where AI can produce them on demand. She outlined three alternatives she now uses in her own courses. First, students can critique AI-generated output, for example, letting a machine draft a lesson plan, then editing it with tracked changes and commentary to demonstrate genuine understanding. Second, debate formats require students to argue against AI in real time, sharpening the ability to think on their feet. Third, role-play simulations, such as a dental student practising difficult patient conversations with an AI playing the patient, allow high-stakes rehearsal without risk to real people.

«If we keep making learning easier, we are damaging resilience and leaving young people, and older ones too, unprepared for the future,» Crompton warned. «The world is making things so easy right now that it is suppressing resilience.»

The hallucination problem

Crompton also raised a subtle but serious risk: as AI models become more accurate, users grow less vigilant. She illustrated this with a now-famous hallucination in which an AI concluded that parachutes are unnecessary for jumping from planes, because the study it drew on involved aircraft close to the ground. «Context matters,» she said. «And the closer AI gets to being right 99.99% of the time, the more we will treat it as an omnipotent god that inevitably lulls our vigilance to sleep.»

She added that AI's performance in research is notably weaker than in other domains, precisely because academic literature sits behind paywalls that models cannot access. This creates systematic blind spots and hallucinations in scientific contexts.

AI agents and the new frontier

Crompton pointed to agentic AI, systems that can be instructed to complete entire tasks autonomously, as the next critical frontier. A Perplexity study found that one of the primary use cases for their platform was students having AI complete entire online courses for them. Her proposed solution is "prompt injection": hidden instructions embedded in course materials that trigger the AI to send an alert email to the instructor, confirming autonomous completion. «Just knowing such traps exist could significantly reduce cheating,» she said, «and it provides solid grounds for academic dismissal.»

Estonia's digital head start

Crompton praised Estonia specifically, calling it one of only two countries, alongside Singapore, she regularly highlights in global forums as models for digital education readiness. «I listened to your prime minister at a UN digital learning week in London last year and thought: you have already done all of this, while others are still scrambling,» she said. «That is a good club to be in.»

On the future of university lecturers, she was reassuring: «Did we get rid of maths teachers after calculators? Not really.» She envisions educators shifting from "sage on the stage" to "guide on the side", deploying AI as a tool while bringing something machines cannot replicate: empathy, personal stories, and genuine care about whether students actually learn.

«AI tools don't care whether we're actually learning,» Crompton said. «They have no feelings. And do you really want to learn from someone who feels nothing?»