Personalised learning breakthrough or the end of thinking for yourself? Both sides, steel-manned, with every claim traced back to a named source. Undated on purpose — this issue gets refreshed as the argument moves.
Eight parts, one cover. Same frame every issue in The Debate; only the words and the two filled grid cells change.
In early 2023, most universities banned ChatGPT. By 2026, most of them had quietly reversed those bans. The tool is in the room regardless of what the policy says — between 74 and 92% of students have already experimented with generative AI for academic work, depending on the survey. The question of whether to allow it has been settled by the students themselves.
What remains unsettled is harder: not whether AI is in the classroom, but how to design classrooms so that it helps the learning rather than replacing it. That is an institutional, pedagogical, and design question, not a detection question. And so far, the institutional response has been fragmented — a patchwork of updated policies, new detection tools, and redesigned assessments, with no coherent framework across sectors or jurisdictions.
This issue lays out the strongest version of each argument, then asks what a genuine design response — rather than a detection arms race — would actually require.
Each side, put as well as its own advocates would put it. Sources named. No figure here is offered as settled fact; each is a claim by a named author, linked in full at the end.
AI-related misconduct grew from 1.6 to 7.5 cases per 1,000 students between 2022 and 2026. UK universities reported nearly 7,000 proven cases in 2023–24, a threefold increase year-on-year — a signal that detection and deterrence are not keeping pace with access.
Pew data from January 2025 found 26% of US teens already use ChatGPT for schoolwork. The tool is not an edge-case experiment — it is a mainstream homework aid for more than one in four students.
Detection bias is a serious equity problem: non-native English speakers face a 61.2% false positive rate from AI detection tools, versus 5.1% for native speakers. Innocent students from diverse backgrounds are disproportionately accused — the detection tools fail the students who need protection most.
Turnitin itself cautions institutions against sole reliance on its AI indicator. The deeper concern: if AI writes the essays, students lose the practice of constructing arguments, organising evidence, and writing under their own steam — precisely the cognitive work that education is meant to build.
— Eyesift
Between 74 and 92% of students have already experimented with generative AI tools for academic purposes, per multiple 2025 surveys. The tool is already in the room regardless of policy. Designing for reality is more productive than designing for a pre-AI classroom that no longer exists.
AI tutors can provide one-to-one feedback at scale — a resource historically available only to students who could afford private tutoring. Well-designed AI tutors give instant, personalised feedback on drafts without a teacher's marking queue, narrowing the gap between well-resourced and under-resourced students.
— Multiple 2025 surveys, cited in The Education Magazine 2026
68% of schools are integrating AI detection, 45% are redesigning assessments, and 58% are updating policies — active institutional adaptation, not paralysis. The majority of institutions are responding, even if the responses are uncoordinated.
AI literacy — understanding what these tools can and cannot do — is itself a learnable, teachable skill that students need for the jobs they will do. Assessment redesign (oral exams, portfolio work, situated tasks) is harder to fake and better at measuring what education claims to measure.
— APA
House rule: any percentage about misconduct or usage is attributed to its author and study. We reference claims; we do not manufacture certainty. Survey figures vary significantly depending on population, question framing, and institution type.
The ban-it-vs-embrace-it debate is largely over. By 2026, most institutions have reversed their 2023 ChatGPT bans. What remains is not a binary choice about whether to allow AI — it is a fragmented, institution-by-institution patchwork of frameworks. The live challenge is design: how to build courses, assessments, and classrooms where AI helps the learning rather than replacing it.
The detection arms race is probably a distraction from the harder design work. Detection tools that produce 61.2% false positive rates for non-native English speakers are not a neutral enforcement mechanism — they are an equity problem embedded in the response to an integrity problem. A more durable answer is assessment redesign: oral exams, portfolio work, in-class writing, tasks that are context-specific enough that outsourcing them is not useful.
The most pointed version of this argument is also the simplest: a test that an AI can pass is a test that was already assessing the wrong thing. If an essay prompt can be answered adequately by a language model with no lived experience of the course, the prompt was never measuring understanding in the first place.
Whether having AI write a first draft that you then edit is still learning is genuinely unresolved. Some argue that writing from scratch is where the thinking happens; others that editing and improving is equally demanding cognitive work. Neither position has settled the question empirically. The answer may differ by student, subject, and what stage of learning they are at.
Most institutions have deployed AI detection tools. Most detection tools perform poorly on non-native English — well enough to catch some cases, badly enough to create serious equity problems for others. Turnitin has publicly cautioned against relying solely on its own AI detection flag. The detection layer is in place; it is not working well enough to be the primary response.
45% of institutions report redesigning assessments in response to AI. This is the more durable response: assessments that are oral, portfolio-based, situated, or in-class are genuinely harder to outsource. The obstacle is time and resource — redesigning an assessment framework takes substantially more institutional effort than deploying a detection tool. It is also the approach that produces better assessments by the criteria education has always claimed to care about.
The least-developed layer is AI literacy: teaching students what these tools can and cannot do, how to use them productively, how to evaluate their outputs critically. This is the skill they will need in the workplace. The APA and multiple 2025 frameworks identify it as teachable. It has not yet been systematically integrated into curricula at any scale. The gap between where this is heading and where most syllabuses currently are is significant.
Every claim above, traced to its author. Read the originals; argue with them, not with us.
The Debate — No. 09 / 09 · AI in Education · Craig Stanley Studio · CS·PRESS · Undated by design, refreshed as the argument moves · One red, used once · All colour pairings meet WCAG 2.1 AA.
Nine debates, each titled and never dated, each refreshed in rotation as the argument moves. Plain English. Both sides put at their strongest. Every claim about the world attributed to its author with a link. No hype, no banned words, no manufactured certainty.
AI and creativity · AI and work · AI as a force for good · AI and young people · AI and ethics · AI and truth · AI and the environment · AI and power.