A conversation with Professor Rose Luckin on where AI genuinely helps learning, where it risks undermining it, and what it means for the educators navigating both.
Key takeaways
- AI in education carries real promise for scaling feedback and surfacing insights that support teachers to know their students better.
- The risk of cognitive offloading is real, but research suggests it is shaped largely by what educators ask students to do, not by the technology itself.
- AI detection tools are unreliable and risk damaging trust between students and schools. Better task and assessment design is a more productive path forward.
- School leaders do not need to act fast on AI. Learning fast and acting more slowly, as Professor Luckin suggests, is the more defensible approach.
- The schools most likely to get AI right are those who begin by defining the purpose they want it to serve, rather than responding to existing tools in use.
There is an experience of Artificial Intelligence (AI) in education that most people in schools are familiar with, by now: the essay produced overnight, the assignment that is technically complete and intellectually empty. This is the version that has generated the most educator anxiety, the most policy responses, and the most calls to do something, anything, quickly.
Professor Rose Luckin, Professor Emerita of Learner-Centred Design at UCL’s Knowledge Lab and founder of Educate Ventures Research, has been at the intersection of AI and education for over 30 years. Her perspective on the current moment is neither alarmed nor reassured. It is, more usefully, precise.
“This moment in time,” she says, “is a moment when we as humans need to get a lot more intelligent.”
That framing repays attention. The question she is asking is not whether AI belongs in schools, but what kind of human thinking it should be in service of, and what happens when that question goes unanswered.
The Difference Between Helpful and Harmful Cognitive Offloading
One of the central concerns circulating in schools at the moment is what researchers describe as cognitive offloading: students using AI to bypass the kind of effortful thinking that produces durable learning. The concern is legitimate, but Luckin’s reading of the evidence adds important nuance to it.
Citing research by Jason Lodge and Leslie Loble, Luckin draws a distinction between beneficial cognitive offloading, where AI handles groundwork and frees students to think at a higher level, and disadvantageous cognitive offloading, where students simply hand thinking over to the tool and receive an output they do not meaningfully understand. The difference, she argues, is not inherent in the technology. Rather, it is in what students are asked to produce.
“When they’re required to demonstrate their own thinking and understanding, they use the AI very differently. A lot of this is down to how we behave with our students, what we ask them to do.”
This is a meaningful reframe for educators. The problem is not AI access. It is task design that only requires students to deliver an output, rather than demonstrate understanding.
As Luckin puts it, drawing on the perspective of James Cook University Vice Chancellor, Simon Biggs: “Generative AI in particular is just shining a light on poor pedagogy, poor assessment, poor curriculum. Things that should have changed, [that] now need to change.”
Where AI Genuinely Helps: Feedback, Insight, and the Teacher’s Role
Luckin is careful to distinguish between the emerging promise of generative AI and the more established body of research on discriminative AI, the AI that predates ChatGPT and has been building an evidence base in education for considerably longer. That research has produced clear findings about what effective AI-assisted feedback looks like: well-timed, appropriately granular, and designed to prompt student thinking rather than replace it.
On the question of AI’s potential to surface insights for teachers, she describes what she calls an intelligence infrastructure: data, AI analysis of that data, and the human educator who interprets what it means for their students. The emphasis on human interpretation is deliberate.
“The really important work happens when you bring the educator, or parent, or peer, into the wider context. It’s not just the technology and the student. It is a wider picture.”
She is direct, too, about where early promise on equity has not materialised as hoped.
“For many years, I believed that AI was the great tool for breaking down barriers, for increasing equality, and that’s not the way it’s playing out, which is a big disappointment. But it doesn’t mean it doesn’t have the potential.”
Luckin’s Beyond the Hype research, and subsequent report Artificial Disadvantage, both found evidence that AI adoption is widening the gap between more and less advantaged communities, rather than closing it. That finding does not mean the potential of AI has vanished, but it does mean that equity cannot be assumed to follow from access alone. She is exploring this finding further in her current research partnership with Education Perfect.
On Detection, Trust and What Educators Can Do Instead
One area where Luckin is unambiguous is AI detection. There is no tool that will reliably and conclusively identify AI-generated work, and the attempt to use one carries real costs. She describes multiple instances in the UK of students being accused of using AI when they had not, and the damage to trust that followed.
“We have to be very, very careful that we don’t inadvertently allow AI to leak into things that should be done by people…Be very careful about the trust that needs to exist between a student and the place where they’re studying.”
Her recommendation is to redirect energy toward assessment and task design that either incorporates AI transparently or asks students to do things the AI cannot do for them. Detection, she suggests, is a dead end. Better design is where the leverage is.
Learning Fast, Acting More Slowly
For school leaders fielding pressure from parents, boards, and governments to act on AI, Luckin offers a useful framework. She describes three distinct ways of thinking about AI in education: as a tool that helps people achieve things, as something that requires educators to rethink what they expect of human intelligence, and as a subject that students and staff need to genuinely understand before use.
Her practical advice builds from there. Define the purpose first. Then, work through four connected dimensions: governance and ethics, use cases, technology and data infrastructure, and staff capability. The order matters because, as Luckin notes, organisations that have an AI policy in place are significantly more likely to be building staff capability in ways that lead to effective use.
Her motto for this topic is “learn fast, act more slowly,” and the distinction in that phrase is crucial. It is not an argument for delay, but for getting the thinking right before the doing, because the schools and systems that moved fastest after the launch of ChatGPT are not necessarily the ones seeing the most benefit.
The vision Luckin holds for what ‘getting it right’ looks like is instructive: AI largely in the background, orchestrated by staff, supporting richer human interaction between teachers and students, and helping build learners who are genuinely capable of continuing to learn throughout their lives. The version she is most concerned about is the inverse: hyper-personalised AI platforms delivering content to students while human relationships are pushed to the margins.
“Education is about building relationships, isn’t it? And that requires human provenance.”
That question is worth returning to as the conversation about AI in schools continues to accelerate. The technology is not going anywhere. The question is what, and who, it is designed to serve.
Hear the full conversation: This article draws on Episode 3 of Well MeasurED, Education Perfect’s interview series exploring what genuinely works in education and EdTech with industry leaders and researchers.
Watch or listen to Dr Jill McGuire and James Santure’s conversation with Professor Rose Luckin, The Truth of AI: What’s Working, What’s Failing, and What’s Next, to hear the research in her own words.
