WHAT THE EXPERTS SAY.... WHY AI DETECTORS ARE DOING
MORE HARM THAN GOOD Comment by REBECCA D’CRUZ, Head of Computer Science and AI Strategy Lead at St Albans School in Hertfordshire
A
I detection tools are unreliable by design. They work by calculating the statistical probability that a given piece of text was machine-generated, analysing patterns in word choice and predictability. The fundamental problem is that these are also the patterns associated with clear, well-structured writing. A student who has thought carefully, expressed themselves concisely and edited their work will produce text that looks, statistically, more like AI output than one who has written hastily and idiosyncratically. The tool cannot distinguish between the two. Research consistently shows high false positive rates, so students are being penalised for work they genuinely produced, while others using AI go undetected with some deliberately introducing spelling and grammatical errors to evade the detector. By using AI detectors in this way, we are building an incentive structure that rewards poor writing over clarity, which is precisely the wrong lesson to embed in assessment.
Most secondary-age pupils are already fluent users of generative AI tools outside of school and they are using them to draft messages, generate ideas, summarise content and navigate everyday decisions. What they often lack, however, is a critical framework for understanding what these tools actually do, where they fail and why the process of
thinking matters independently of the output. As a result, schools are in the uncomfortable position of being slower to adapt to AI than their students. When institutions respond by banning tools that pupils use daily beyond the school gates, they lose credibility and, more importantly, they lose the opportunity to shape how young people engage with AI in a thoughtful and discerning way.
Bypassing a struggle means bypassing learning When it comes to examinations, our assessment structures were not designed for a world where generative AI exists. For instance, a GCSE essay question asking students to analyse a text, construct an argument or evaluate a historical event was designed for a world where the cognitive effort required to produce a response was itself the point. Generative AI collapses that effort. The structural problem is not that students are lazy, because most are not, but that the assessment does not distinguish between a student who has thought carefully and one who has outsourced that thinking. Younger students in particular often lack the metacognitive maturity to recognise that bypassing the struggle is bypassing the learning.
Moving forward, we have no choice but to redesign the assessment process, moving beyond “did a human write this?” towards process portfolios, annotated drafts and dialogic tasks that foreground reasoning and development. In the near term, the most robust approach would likely involve a foregrounding process rather than product. For example, a student who submits an annotated draft which showcases their initial
July/August 2026
thinking, the questions they asked, the revisions they made and why, is demonstrating something that AI cannot fabricate on their behalf. Dialogic assessment, where a student discusses their work verbally with a teacher, is similarly resistant to substitution.
Longer term, examination boards will need to reckon with this process seriously. Controlled assessment conditions remain the most defensible environment, but they test a narrow band of performance. The honest answer here is that we might not have the certainty or guarantee that a final product was written by a human; but we can, however, design tasks where the human thinking is the visible and assessable artefact.
Using AI with transparency and sound judgement Genuinely effective AI integration means teaching students to use these tools critically, transparently and with sound judgment. Critical use means understanding that a generative AI output is a probabilistic construction, not a reliable source of truth, so students need to interrogate it as they would any text of uncertain origin. Transparent use means being explicit about when and how AI has been used in the production of a piece of work, a habit of academic honesty we should be building in young people from an early age.
Judgment means knowing when AI adds value and when
it hinders learning and progress. Using it to stress-test an argument or generate alternative perspectives is quite different from using it to avoid thinking something through. These are teachable skills, and schools that treat them as such will serve their students far better than those that simply attempt to prohibit.
The question is, is AI literacy really a threat to academic rigour or should we be viewing it in a different way? Whatever one’s view of generative AI, the question of whether schools should engage with it is already settled: students are using these tools now, inside and outside of the classroom. The framing of AI literacy as a threat to rigour assumes that rigour is a fixed property of existing assessment structures, rather than a commitment to genuine understanding and intellectual development, which should never be the case.
Protecting the values of rigorous education
Academic rigour means ensuring students can think, reason, evaluate and communicate and none of those capacities are diminished by learning to use AI tools well. What would diminish them, however, is allowing AI to replace those processes unexamined. The real threat to rigour is not AI literacy but AI invisibility: a culture where students use these tools unreflectively, without understanding their limitations or their own responsibility as independent thinkers. Teaching AI literacy is not a concession to the technology. It is really about how we protect the values that rigorous education has always been trying to serve.
www.education-today.co.uk 31
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40
