AI and education: We need to know where this sudden marriage is heading

What does “happily ever after” look like for artificial intelligence (AI) and education? No one seems entirely sure.

We’ve moved past the question of whether they should be together. Largely because AI didn’t wait for an answer. But making the relationship work, on terms that actually serve human development, is proving to be a Gordian knot.

The AI world keeps rolling out new tools that might sound relevant. ChatGPT is reportedly experimenting with a “Study Together” feature that prompts users to ask more questions.

But like many countries, Singapore’s approach has been more deliberative than gung-ho. Education Minister Desmond Lee has emphasised that artificial intelligence must enable learning, not supplant it. The focus has been on age-appropriate classroom use, AI literacy, and ensuring that curiosity and social-emotional development aren’t crowded out by code and prompts.

It gives us a snapshot of where the AI-education marriage stands – for now. Caution is no bad thing, especially as we continue to reckon with the digital whiplash that hit a generation of young people in recent years. But there is a deeper, more fundamental dilemma that even the experts are grappling with: What kind of education do we want AI to serve?

At a time when education’s promise of a stable job seems wobbly, clarity over AI’s role matters at every level of learning. If machines are replacing jobs or reinventing them, we need informed, urgent conversations about how central we dare to make it so that learning remains relevant for the future.

Good. Bad. Good. Bad.

In a May 2023 TED Talk, Mr Sal Khan – founder and CEO of Khan Academy – introduced Khanmigo to the audience. This is an AI-powered “super tutor” chatbot designed, as he insists, to assist teachers, not replace them. It provides real-time help for students and live insights for teachers, flagging when a learner is stuck or disengaged. The idea is to free up educators for something AI chatbots still can’t offer: meaningful human connection.

It was a glimpse of how personalised, one-on-one tutoring could become accessible to all – not just the lucky or wealthy few. And that, he said, could solve education’s well-documented 2-sigma problem – that individual, tailored tutoring yields better outcomes than traditional classrooms.

Not everyone is sold. Some experts suggest too much is being made of AI’s potential and promise, pointing to a deeper question: Is the purpose of education simply to make mastery easier? To generate quick answers?

Professor Rose Luckin of University College London put it plainly in a LinkedIn post. “Are we building students’ capacity to thrive in an AI-integrated world, or are we simply making traditional education more efficient?” She was reflecting on a keynote address by Mr Khan at the EduTech forum in Australia in June 2025, where he was again championing AI tutoring.

Prof Luckin, whose work involves ethical AI for education, warned that we may be retracing old ground, focusing on content delivery and mastery at the expense of cultivating the ability to “learn how to learn”.

Educators I spoke to seem split. “Why should I learn this if AI can just do it for me?” That’s the refrain one hears more and more from students, said one lecturer. Another, more sanguine, tells me how the technology has helped him in designing lessons or rubrics which he can then build on.

Studies and surveys haven’t exactly settled the debate.

A Pew survey conducted in 2023 found that a majority of 12th grade teachers were uncertain or found downsides to the use of AI tools in teaching. Cue the flip side. A recent Gallup survey of over 2,000 US public schools found that almost 60 per cent of teachers using AI every week saved six hours of work.

Some findings are more depressing than others.

At MIT’s Media Lab, researchers ran a controlled study involving essay writing. Participants were split into three groups: one went it alone, another used Google, and a third relied on ChatGPT. The results were as sobering as they were predictable: ChatGPT users showed the lowest cognitive engagement and consistently underperformed on neural, linguistic and behavioural measures. Over time, they became less curious and more copy-paste.

The study was small (54 participants), but lead researcher Natalia Kosmyna found the results concerning enough to publicise. “Developing brains are at the highest risk,” she warned.

Like it or not, AI is now stitched into the daily business of schools through ChatGPT and its offshoots – some to cut corners, some to spark ideas. But while studies and surveys provide more insights, they have us on the edge. It feels like disruption without direction – hardly reassuring for anyone trying to make sense of education, let alone plan a career.

For educators, it can feel like drawing a line in the sand as they grapple with setting limits on AI use in assignments and assessments. The recent NTU saga – where students were accused of academic fraud for using Gen AI tools in an assessment – reveals how hard it is to adjudicate intent, let alone enforce boundaries in an age of AI’s omnipresence. 

Do we add a new layer of rubrics to determine competency? Or tear up the curriculum and start again?

Finding a sweet spot

At this stage, experts and policymakers must confront not just those questions, but also the purpose of education itself. Do we need to redefine what it means to be educated? How do we reconcile the age-old goal of nurturing the ability to learn with today’s accelerating race for instant answers?

Part of the answer lies in how we intend to adopt AI in the first place. Associate Professor Tan Seng Chee from the National Institute of Education (NIE) researches on education technology for schools. He laid out some considerations: accessibility of Gen AI tools, the ethical issues surrounding the use of AI, training for educators and inclusivity in the design of AI applications. But he also added that the speed at which AI is advancing complicates this.

“We are witnessing the introduction of more powerful generative AI tools, and some researchers are saying that prompting skills are no longer that important. What this means is that even if we develop some strategies with current AI tools, we might need to adapt quickly when new tools are available,” he said.

But this isn’t just about new tools. As AI evolves, it risks streamlining away the very struggle that makes learning meaningful. It’s another crossroads for the system of teaching and learning that we have become so used to.

In a podcast, Professor Rebecca Winthrop, co-author of The Disengaged Teen, offers a sharp warning. AI’s greatest pitfall, she argues, is its removal of friction from learning. Its tendency to smooth over challenges could deep-six the “learning pit” which essentially is the struggle and perseverance essential to gaining mastery. No resilience, no growth.

Prof Winthrop, a global education expert from Brookings Institution, insists that children must build what she calls “explorer muscles”, the resilience to dive into complexity and emerge with understanding, not bypass that experience entirely.

She shared a story of a student gaming the system. A high-schooler split an essay prompt across multiple AI models, recombined the pieces, then ran it through an anti-plagiarism checker. A shortcut savant.

“No matter what, kids will find a way. We cannot out-manoeuvre them with technology,” she said. Do we then risk turning education into a process of assembling answers, rather than cultivating intellectual grit and creativity?

Engagement is often a stronger anchor than intelligence for long-term learning. Some blame AI for its erosion, while others showcase how it can be leveraged for the better. Culprit or catalyst?

This is where the real tension lives. There’s probably a sweet spot that exists between blind embrace and outright rejection – where AI’s promise can be harnessed without letting it rewrite the script of learning itself. Finding it would require a clearer understanding of the technology itself and how we chart the course, even as it rapidly evolves.

Research and development, as Prof Tan tells me, plays a critical role: “On one hand, we need to monitor the long-term impact of generative AI on learning and learners; on the other hand, with the rapid advancement in AI technologies, we need to have a rapid research cycle that can adapt to these changes to harness the power of AI.”

Maybe, we are getting better at learning to adapt as AI unfolds. Even as we do that, there’s a fear that we may surrender too easily to the trade-off between ease and effort. The next generation may master the art of learning quickly without learning how to think deeply.

At that stage, AI didn’t beat us... We handed it the win.

Read the original article here. 

Source: The Straits Times © SPH Media Limited. Permission required for reproduction.