Is AI Making Our Children Stupid? The Case for Hitting Pause on EdTech

This week Broken Custodians explores the rapid proliferation of AI in education, the evidence that it undermines learning, and the risks of conducting a mass EdTech experiment on children.

This weekend Education Secretary, Bridget Phillipson MP, has acknowledged in an article for The Telegraph that schools are having to navigate a ‘Wild West’ of EdTech products. In this week's episode of Broken Custodians we examine the government's increasingly enthusiastic embrace of AI in the schools system, and consider whether, in the absence of a meaningful evidence base, we are sleepwalking into a crisis for children's wellbeing, learning and safety.

We are joined by education policy analyst and former advisor to the Department for Education, Tom Richmond, to ask a question that too few in government seem to have contemplated: is AI in our school system a net positive for children’s learning?

Drawing on his research for a paper published recently by the Social Market Foundation, Tom explains that while the political enthusiasm for generative AI in education is unmistakable, the evidence base to justify it is so far almost entirely absent. Worse, the evidence to date suggests the use of AI tools could have a seriously negative impact on children’s learning and knowledge retention.

At the heart of Tom’s argument is a deceptively simple principle from the science of learning: learning happens when you think hard. Tools like ChatGPT and Claude don’t make the process of thinking faster, they bypass the thinking altogether. Tom recounts a landmark randomised control trial which found that students who used AI chatbots to help solve maths problems in class outperformed their peers by 127%, but when the chatbots were removed for the final exam, those same students performed worse than the students who had not had access to AI: in other words, AI use appeared to have been net negative to learning and knowledge retention.

Tom also describes a brain-scanning study which found that students writing essays with the help of an LLM tool exhibited the least brain activity of any group and, strikingly, at the end of the test not one of those students could recall a single line of the essay they had just ‘written’.

We ponder whether, despite all of these reservations, we might expect an AI teacher to deliver better educational outcomes for pupils than a poor quality human teacher. And we question the ethics of deploying AI and EdTech across the school system before minimum safety and efficacy standards have been established. Tom’s key takeaway is a warning that parents, teachers, and policymakers are dangerously close to losing sight of the distinction between performance and learning, and that this could prove disastrous for the skills and attainment of a generation of children.

Here is the full episode which you can also find on YouTube, Spotify and Apple Podcasts…

Broken Custodians is an UsForThem podcast and Substack https://www.youtube.com/@Broken.Custodians