The Missing Middle- Why AI in Education Needs Guardrails, Not Hype or Panic

Abstract

Artificial intelligence is being sold to schools, parents, and students as a near-universal good — a tool that personalizes learning, saves teachers time, and helps struggling students catch up. The marketing is confident. The evidence, so far, is not. That gap between promise and proof is the real story here, and it matters more than a simple "AI good" or "AI bad" verdict.
A 2026 case study examining AI tool use in an asynchronous university course found no statistically significant difference in final grades between students who used AI tools and those who didn't [1]. What did predict performance, across both groups, was active engagement with the course, not which tools students used [1]. In other words, the technology itself was not the variable that mattered, which is an uncomfortable finding for an industry whose pitch rests almost entirely on the technology being the variable that matters. This is not an isolated result; it is a symptom of a wider pattern that researchers inside the field of educational technology, not outside critics, have been naming for several years. Neil Selwyn, a leading scholar of education technology, has spent much of the last few years cataloguing the ways AI's educational benefits get overstated. In 2022 he outlined several concerns with the trajectory of AI in schools, warning against uncritical acceptance of industry claims and pointing to the social harms, ideological assumptions, and environmental costs that tend to be left out of the pitch [2]. In a 2024 follow-up, Selwyn argued that the sheer volume of hyperbole surrounding AI has made it harder to think clearly about its actual, lasting educational implications, and cautioned that uncontrolled reliance on these tools risks reducing authentic learning experiences and genuine cognitive engagement [3]. A separate 2024 study built and validated a model of AI's underdiscussed downsides in education, drawing on 56 reviewed studies and a survey of 260 university participants; it confirmed concerns spanning reduced human connection, data privacy and security risks, algorithmic bias, weakened critical thinking, and unequal access, and found these problems tend to compound each other rather than occur in isolation [4]. None of this is fringe skepticism. It is coming from the people who study education technology for a living.