A
secondary school teacher in Leeds piloted ChatGPT use with her Year 10
English class in 2024, asking students to use it to generate a first draft of
a persuasive essay, then critique and revise the AI output. The results were,
by her account, the most engaged and productive lesson on persuasive writing
she had taught in fifteen years. Students who normally struggled to begin
writing had a starting point to react to. Students who normally wrote
competently had their assumptions challenged by an AI draft that was fluent
but superficial. The exercise taught critical reading, analytical writing,
and AI literacy simultaneously. It was also, she noted, completely outside
the guidance her school had given on AI use, which was to ban it entirely.

The divergence between that teacher’s experience and her school’s
policy is a microcosm of the state of AI in education globally. At the policy
level, most educational institutions have responded to large language models
primarily through prohibition or restriction, driven by legitimate concerns
about academic integrity that are real but incompletely theorised. At the
classroom level, teachers and students are navigating a technology that is
already deeply embedded in how young people learn, write, and seek
information, with or without institutional approval. The institutions that
are most effectively navigating this moment are those developing thoughtful
pedagogical frameworks for AI use rather than either blanket prohibition or
uncritical adoption.

The Academic Integrity Problem

The concern that drove most initial institutional responses to
LLMs was, and remains, academic dishonesty. Students who can produce a
well-structured essay using ChatGPT in minutes have an unfair advantage over
those who do not, and the assessments used to evaluate learning are
undermined when they can be completed without the learning they are designed
to measure. AI detection tools, including GPTZero and Turnitin’s AI detection
features, have been deployed by thousands of educational institutions in
response, but their reliability is significantly lower than their marketing
suggests. Studies from Stanford and elsewhere have found false positive rates
high enough to risk wrongly accusing honest students, and false negative
rates that mean motivated students using AI can evade detection with modest
effort. Relying on AI detection as a primary academic integrity response is a
problematic strategy that creates injustice risks alongside its genuine
deterrent value.

The more substantive response to academic integrity concerns is
redesigning assessment rather than policing tool use. Assessments that
require personal knowledge, in-person demonstration, iterative development
with documentation, or specific local context that an LLM cannot access are
significantly more robust to AI-assisted dishonesty than standard essay
formats. The RSA’s
education programme has published guidance on AI-resistant assessment design
that is being adopted by universities and schools seeking a more sustainable
integrity response than detection-based approaches can
provide.

The Learning Opportunity

The pedagogical case for thoughtful LLM integration in education
is stronger than the institutional reluctance to engage with it suggests.
Research from MIT’s Education Lab and the OECD’s Centre for Educational
Research and Innovation has found that students who learn with AI assistance
on appropriately designed tasks develop stronger metacognitive skills than
those who work without it, because the process of evaluating, critiquing, and
revising AI output requires and builds the kind of reflective thinking about
quality and reasoning that is central to deep learning. The Leeds teacher’s
experience is not anecdotal; it reflects a pattern of outcomes in
AI-integrated pedagogical experiments that is increasingly
well-documented.

Personalised learning is a longer-term aspiration that LLMs are
beginning to make practically achievable. Intelligent tutoring systems that
adapt to individual student performance have shown consistent positive
effects in controlled studies, and LLMs provide a significantly more flexible
and natural interface for these systems than the rigid structured software
that preceded them. Khan Academy’s Khanmigo, which uses GPT-4 to provide
Socratic tutoring dialogue to students, has shown promising early results in
increasing student engagement and learning outcomes in mathematics, with
published evidence from randomised trials that is more rigorous than most
edtech evaluation.

Equity and Access Concerns

The equity implications of AI in education are multidirectional
and incompletely understood. On one hand, access to high-quality AI tutoring
through tools like Khan Academy’s Khanmigo could reduce educational
inequality by giving students without access to private tutoring a form of
personalised learning support that was previously unavailable to them. On the
other hand, if AI tools are integrated into educational practice in ways that
benefit students with better digital access, more supportive home
environments, and greater AI literacy, existing inequalities could be
amplified rather than reduced. Research from the National Foundation for
Educational Research has specifically examined the equity
implications of AI in UK schools, finding that the benefits of AI educational
tools are accruing disproportionately to students in more advantaged schools,
a finding that has significant implications for how AI education policy
should be designed.

What This Means for You

For students, the practical question is how to engage with AI
tools in ways that develop genuine capability rather than substituting for
it. Using LLMs to generate content and submit it as your own work is both
academically dishonest and self-defeating; the skills that education is
designed to build are the skills that the labour market will value throughout
your working life. Using LLMs as a thinking partner, a source of feedback on
drafts, a tool for exploring ideas and identifying gaps in your
understanding, is a genuinely valuable learning practice that prepares you
for an AI-integrated professional environment. For parents, the relevant
question is whether your children’s schools have thoughtful AI policies that
engage with the pedagogical opportunities rather than simply prohibiting a
tool that students are already using outside the classroom. The international
evidence on LLMs in education is broader and more encouraging than the UK
policy response currently reflects. Countries including Estonia, Finland, and
Singapore have developed national AI in education frameworks that treat LLMs
as pedagogical tools requiring thoughtful integration rather than threats
requiring prohibition, and the early evidence from these systems suggests
meaningful improvements in learning outcomes for students whose teachers are
trained to use AI effectively. UK education policy is lagging behind these
international examples in ways that will have consequences for the AI
readiness of the next generation of students and workers. The Department for
Education’s AI in education strategy, published in 2023, acknowledged the
potential of AI tools but has not been followed by the investment in teacher
training, curriculum development, and equity infrastructure needed to realise
it. Closing this gap is an urgent educational and economic priority. For
related analysis, see our coverage of AI
and graduate employment and the
AI skills economy.

The teacher professional development challenge is significant and
underinvested. Effective AI integration in education requires teachers who
understand what LLMs can and cannot do, who have the pedagogical knowledge to
design AI-integrated learning experiences that build genuine skill, and who
have the confidence to navigate institutional policies that may lag behind
what good practice requires. Current teacher training programmes in most UK
initial teacher education providers have not yet systematically integrated AI
pedagogy, meaning that newly qualified teachers are entering classrooms
equipped with AI tools but limited frameworks for using them responsibly and
effectively. The Teaching Futures initiative at UCL and similar programmes at
several other universities are developing AI pedagogy training that could
address this gap, but scaling these programmes to reach the entire teaching
profession requires investment and policy priority that the Department for
Education has not yet committed. The long-term quality of AI-integrated
education in UK schools depends significantly on decisions about teacher
professional development being made, or not made, in the next two to three
years.

About the Author

Stuart Kerr is a technology correspondent at LiveAIWire, covering
artificial intelligence, digital innovation, and the social impact of
emerging technologies. Follow LiveAIWire for daily analysis at liveaiwire.com.