AI in Chinese schools: a national plan, but learning outcomes remain an open question
Artificial intelligence is entering Chinese schools in very different ways: building applications, supporting study and addressing teacher shortages are among the cases described in a Reuters investigation published on 16 November and read on Euronext.
The issue is not just how much technology reaches classrooms, but what work remains for students: Reuters describes a sweeping, uneven expansion without settled agreement on how AI literacy should be taught or measured.
## A goal for 2030, not an outcome already achieved
The “AI + education” action plan from China’s education ministry and four other departments targets deep integration by 2030 and the inclusion of AI education in local curricula.
The document is dated 2 April 2026 and was published on the ministry’s website on 10 April: it provides context for the new investigation, rather than a measure announced today.
It calls for teacher training, educational applications and security rules for algorithms, data and systems, including risks of fraud, academic misconduct and privacy breaches.
These are programme commitments, not certification that every school has already met the same standards.
## Adoption does not establish effectiveness
Reuters says it reviewed more than 100 public procurement documents, identifying commitments of at least US$31 million over the past year for AI classroom equipment, training and teaching platforms.
That is the scope documented by the investigation, not an official national spending total.
In rural Anhui, a headteacher describes personalised tablet lessons and agents that ask questions when pupils struggle.
The university-admission figure claimed by that headteacher is not used here as evidence of effectiveness: Reuters says it could not verify it independently.
## A useful check is what students can do without an assistant
The investigation also reports studies with mixed signals about immediate results and independent skills, explicitly noting that the observed associations do not establish causation.
One interviewed mother describes a household rule: try a problem for fifteen minutes before seeking help, then attempt it without AI again several days later.
This is an individual experience, not a validated educational prescription.
The question worth bringing to the debate is straightforward: are we measuring the quality of the answer produced, or the capability retained by the learner?
