By Professor Mark Mon- Williams
Technology is placing our children at immense risk. There are the well documented problems associated with screen time and social media. But there are other assaults on the rights of children that need our urgent attention.
First, AI and associated technologies are changing the educational landscape at a breathtaking pace. The technology has fantastic potential to accelerate children’s learning but carries the risk of increasing the educational inequity that already blights the UK.
Second, the “if you are not paying for it then you are the product” adage highlights that companies need data to train a product for monetisation. This raises serious concerns about the level of data protection provided when children and young people engage with a technological product.
These concerns are amplified when we consider “educational technology”. Schools and educationalists have an understandable desire to accelerate learning through new technological products. But this raises fundamental questions for policymakers about how to check that a new product is safe, effective, and protects children’s rights.
Moreover, these questions need to be answered in an age of information warfare. We are in danger of drowning in a swamp of noise. Every new AI announcement brings another claim about what technology will do for education. Research papers compete for attention with product evaluations, personal testimony, and confident predictions.
For policymakers, the challenge is how to make sense of the evidence arriving on their desks. This is why I welcome the Department for Education publishing a Framework containing eight principles for evidence-informed policymaking in fast-changing technology contexts.
The central argument in the publication is that we should combine different sources of evidence when considering new technologies, assess their limitations, and update our understanding as new findings emerge (a process labelled as Bayesian Evidence Accumulation).
This practical proposal raises a fundamental epistemological question: how can government know enough about the world to change it for the better? Every policy rests on a ‘world model’: an account, explicit or otherwise, of how things work and what an intervention might achieve. Effective policymaking requires that the model tracks reality closely enough to guide action, while acknowledging where our understanding remains uncertain.
A world model therefore requires more than running a study. It must connect observations to explanations. If an AI tool appears to improve attainment, did the technology cause the improvement, or did the adopting schools already have advantages that would explain the findings? Would the same effect occur elsewhere? Different explanations imply different policies. Evidence becomes useful when it helps us distinguish between different possible explanations.
Bayesian reasoning provides a discipline for this task. We begin with existing knowledge and assumptions, then ask how much a new observation should change our confidence in competing explanations. A finding that would be unsurprising under several explanations tells us less than one that clearly distinguishes between them.
This approach is essential because policymakers cannot pause the world while they commission a randomised controlled trial. Schools and families still need guidance, children continue using the technology, and the products may change substantially before the results arrive. RCTs remain powerful tools for testing causal claims, but waiting for one is itself a policy choice with consequences. Government must therefore act on the best available evidence while continuing to test and revise its understanding.
There is a deeper danger. We can use a study to update our confidence within a flawed model without recognising that the model itself needs to change. If we equate time spent using a product with learning, collecting more usage data may make us more confident while leaving the educational question unanswered. We must question what we measure, where experiences are missing, and whether our assumptions allow us to recognise failure.
This is why the Framework’s emphasis on context, equity, and continuing evaluation matters. Government needs to understand why a technology helps, under which conditions, and for which children. Average improvement can conceal worsening outcomes for a particular group. Apparent disagreement may reveal differences in implementation or populations. Preserving those differences helps us build a more accurate model whilst smoothing can produce misleading reassurance.
An accurate model cannot, however, decide what government ought to value. Evidence can inform judgements about likely benefits and harms but that doesn’t determine which risks are acceptable. Children’s rights must shape those decisions. As the Framework recognises, credible risks of serious or irreversible harm can justify precautionary restrictions. Uncertainty has consequences, and those consequences differ depending on whether we mistakenly adopt a harmful tool or delay a beneficial one.
Maintaining a useful world model is therefore a societal responsibility. The publication calls for infrastructure that supports continuing evidence accumulation and makes findings accessible. I would go further: government needs a revisable account of its important claims, the evidence supporting them, the competing explanations and the observations that would trigger reconsideration. Independent scrutiny matters because the organisations selling products should not control the evidence by which public claims are judged.
The Department for Education are well positioned in this regard. They have established a Scientific Advisory Council that helps connect government with the relevant academic expertise. This is evolving as a powerful partnership: policy colleagues bring the questions that require decisions, while researchers help establish what the evidence can tell us. We must all be able to question the assumptions in this process. A shared world model provides a dependable starting point for discussion while keeping consequential disagreements visible and open to challenge.
On the topic of challenge – this process arguably challenges the way that my academic community currently operate. I have become convinced that publishing a paper cannot constitute completing our community’s academic responsibilities. We must explain how our findings fit with existing knowledge and what would change our conclusions. A busy colleague in DfE should not have to reconstruct an entire academic debate to understand whether a finding is relevant. Children, families, and teachers also need ways to challenge a model that misrepresents their experiences.
Safeguarding children’s digital rights requires government to understand the world it is changing, recognise when its understanding is wrong, and act on what it learns. It requires academics to support them in their endeavours. We must all play our role in developing the necessary understanding as we seek to build a world that works for all children and young people.