GSE set out a position on artificial intelligence in 2023, when most schools were still deciding whether to ban it. The position was that humans belong in the AI loop: that people identify the patterns, assign the meaning and make the decisions that matter, and that a machine supports that work without replacing it. Three years on GSE holds the same position, and the cost of getting it wrong has risen.
In 2023 the argument was mostly about opportunity, meaning personalised practice for students and less marking and administration for teachers. Those benefits were real and some of them have arrived. The part nobody was writing about was what happens to a school when the evidence it relies on stops being evidence.
Students adopted it faster than schools did
The adoption numbers are no longer in dispute, and across American schools the share of students using AI for homework rose from just under half in May 2025 to nearly two thirds by December of the same year, with most of the growth among middle and high school students. That is a settled change in how homework gets done, and it happened inside a single academic year.
Most schools responded with a policy document, and students went on using it anyway. The gap between what a school’s AI policy says and what happens at a kitchen table on a Tuesday evening is now the widest gap in education, and pretending otherwise is the position many schools have quietly adopted.
The real problem is evidence
Framing this as academic dishonesty leads schools to the wrong response, which is detection. Detection does not work reliably, and it turns teachers into investigators in a dispute the school cannot prove either way.
The real problem is narrower and more serious than dishonesty. A school’s entire account of whether a child is learning rests on work the child produces. That work is essays, problem sets, projects and coursework, and when it can be generated in seconds, the school loses its instrument. The school can no longer tell the difference between a student who has understood something and one who has not, and a school that cannot tell has lost sight of the only thing it exists to do.
The universities reached this conclusion first because their exposure was greater. Princeton abandoned a hundred and thirty year old tradition of unproctored examinations in 2026. Across American and British institutions, supervised halls, handwritten work and oral examinations have come back, and they have come back because those formats still produce evidence.
Schools have to decide what now counts as evidence
The schools handling this well have stopped arguing about tools and started rebuilding what they accept as proof of learning.
That means more of the work happening where a teacher can see it, starting with writing done in class where the process is visible. Oral explanation matters more than it did, because a student has to defend an argument without assistance and a teacher learns more in four minutes than a polished essay reveals in four pages. Schools are also keeping drafts and working alongside the finished piece, and setting more practical work that cannot be generated at all.
None of this is new, and all of it was normal before the technology made it inconvenient. The schools rediscovering these formats are not being conservative, they are being accurate about what they can still measure.
The second half of the job is harder and matters more. A student leaving school in 2030 will work alongside these systems for their entire career. A school that only restricts is preparing them badly. The skill to teach is judgement: when to use it, when not to, how to tell whether the output is right, and what it means to be responsible for work a machine helped produce. That skill cannot be taught by a school that has banned the tool, and it cannot be taught by one that has surrendered to it. GSE has written on the skills educational leaders need in the age of AI.
What this has looked like in GSE schools
These tools have been used extensively across the schools GSE manages, and almost all of the gain has landed on the teacher’s side of the work. Operations run with less effort for more effect, and planning improved once curriculum standards could be tracked without somebody maintaining it by hand, which makes coverage and gaps far easier to see during the year instead of afterwards.
The clearest benefit is one that rarely appears in the writing on this subject. International schools employ many teachers whose first language is not English, and for those colleagues written communication has become faster and considerably better. Reports, emails to parents and documentation now read the way the teacher intended them to read, and the effort that used to go into producing them goes somewhere more useful.
One boundary matters as much as the gains, which is that safeguarding is kept away from these tools entirely and the records that go with it stay where they have always been. The efficiencies also reduce cost, which is the part an owner notices, although no school should assume the saving arrives on its own.
The effect GSE expects to matter most is slower to appear. Teachers who have more balance teach better, and a school where staff are not exhausted is a better place to learn. Whether that shows up as improved retention is too early to judge, and it is the number GSE is watching.
Underneath all of it sits a decision about culture. Teachers in GSE schools use these tools openly and are not asked to conceal it. They experiment, and they tailor both material and approach to the class in front of them. A school that drives its teachers’ use of AI underground loses the ability to see what is working, which is the same loss it suffers when students hide theirs.
What humans in the loop means in practice
The phrase is easy to agree with and just as easy to leave as a slogan, and inside a school it carries concrete obligations.
One is that a person makes the instructional decision. A system can indicate that a student is struggling with a concept. It cannot know that the same student’s family is going through a separation, and that is the reason the teacher decides what happens next.
Another is that a person is accountable for the output. Where AI drafts a report to parents, a plan or a set of comments, a named human being has read it and takes responsibility for it, so that nothing leaves the school unchecked.
The last is that a person owns the relationship. The thing that makes a child work harder than they intended is almost never a piece of software. It is an adult whose opinion they care about. Technology makes learning visible; a teacher is what makes it matter.
What an owner or a board should be asking
For owners and boards this is a governance question as much as a pedagogical one, and the questions worth asking are short.
What has changed about how this school assesses students since 2023, and if nothing has changed, why not. What proportion of grades now rests on work produced without supervision. What is the school’s actual practice on AI use, as distinct from its policy. What are teachers using it for, and has it given them back any time. And whether anyone has checked that the money spent on AI tools has improved anything that can be measured. These belong alongside the other indicators an owner should see each month, set out in the school dashboard.
The last question is the one that gets skipped. The education technology industry has always been good at selling to schools, and AI has made it better at it. A school can spend a great deal on systems that produce no measurable improvement, which was true of interactive whiteboards and tablets before it was true of this.
The test has not changed
The question GSE has always put to any technology still applies. Will this increase learning and achievement? The question is harder to answer now than it was for a whiteboard, because for AI the answer depends entirely on what the school does around it.
It earns its place when it helps a teacher plan or gives a student practice with feedback, and it does harm when it produces work nobody has thought about, because that removes the only thing that made the work worth setting. The difference lies in whether a human stayed in the loop.
Background reading
GSE’s earlier writing on artificial intelligence, published in 2023 as the first generation of these tools reached schools:
- AI in Education: Embracing a Human-Centred Approach. Where the humans in the loop position was first set out.
- Empowering Education: Eight Recommendations for Harnessing AI. The fullest of the 2023 pieces.
- AI and Teaching: Empowering Educators in the Digital Age. Written for teachers.
- Unleashing the Potential of AI in International Education. Written for leaders and investors.
- Artificial Intelligence in Education. The first of them.
Read next
- The School Dashboard: What an Owner Should See Every Month. The indicators that sit alongside these questions.
- Vision and Mission Integrity. Testing whether a school’s stated purpose shows up in what it does.
- 10 Skills Every Educational Leader Needs in the Age of AI.
Part of GSE’s What We Believe collection.
To talk about your school project, reach us through the contact page.