As AI Enters Education, Sometimes the Gas, Sometimes the Brake
Illustration · not a service recordAs AI enters education, I increasingly feel that we may need to step on the gas at some times and on the brake at others.
Outside the formal education system, we should in fact be very aggressive in exploring just how far AI can push learning.
We should even let a small number of capable and highly motivated young people run very fast. They may be only senior high school students, or even junior high school students, and already they are not just “using AI” but beginning to understand AI and create AI.
Inside the formal education system, however, I think we need to be more cautious.
That is because schools are dealing not with a few highly self-directed students, but with almost all children.
What effect does AI actually have on the long-term learning of most children? Will it lead to cognitive outsourcing? Will it create emotional dependence? At what age, in which subjects and at which stage of learning is it appropriate to use? And what supporting measures do teachers need?
More and more research is emerging, but we do not yet have sufficiently complete answers.
So I do not think we should be pursuing a “100% AI adoption rate” right now, and still less should we require every teacher and every child to learn with AI.
We can let those who want to run go first, and let them run very fast; but we should not push everyone into the water together.
In a non-totalitarian society, any innovation has to travel the Adoption Curve:
Innovators → Early Adopters → Early Majority → Late Majority → Laggards
Each group has its own pace. Innovation that can truly scale does not skip these stages; it builds up evidence, cases and trust one step at a time.
So my current view of AI × education really sits between two extremes:
Banning AI from education altogether is a head-on collision with the direction of the times;
Embracing AI wholesale out of excessive optimism is also very likely doomed to disappointment.
Because the truly hard part has never been “whether we have AI.”
The truly hard part is this: Do we understand learning? Do we have good instructional design? Do teachers have enough supporting measures?
This is also why Junyi Academy has chosen to do several things for now:
1. Let young people really come in. We have senior high school interns, and we hope the next generation will not only be users of AI but will have the chance to become creators.
2. Step into the international frontier. We take part in international communities on AI education, AI ethics, AI × brain science and more, to understand what the world is now worried about, what it is testing, and where it is breaking through.
3. Treat “effective” as a question that needs to be verified, not as a matter of faith. We work with national-level education research institutions to test in practice whether the education AI we develop really improves learning, including learning outcomes and the quality of students’ questions.
4. Learning First, not AI First. We do not design products for the sake of “having used AI.” We first ask: How do students learn to begin with? Where do they get stuck? For example, students watch videos and do exercises, and only when they really run into difficulty do we let AI step in. AI is a tool within the learning process, not learning itself.
5. Pioneer First, not everyone at once. Whether in elementary, junior high or senior high school, we look first for teachers and schools that are willing to try of their own accord. We first work with Pioneers to produce cases that truly work, and then spread outward step by step from the Early Adopters.
6. Do not let any single party decide the future of education alone. We bring in different stakeholders, including industry, government, academia, research institutions and frontline educators, to think together about how to find, between the ideal and the practical, the healthiest path for AI to enter education.
And this may also be a very important role for NPOs in the AI era: to be the R&D department for public affairs.
Companies have to answer to the market, governments have to answer to the whole population and to political accountability, and schools have to keep running normally every day. By comparison, NPOs have more room to take on experiments that are important but not yet certain to succeed.
We can allow ourselves to fail, but we must be serious about drawing evidence from failure.
The point is not to rush to prove that AI definitely works, but to work hard at finding out under what conditions, for whom and with what methods AI can really help learning.
Once these patterns gradually become clear, and teachers, students and schools find it useful and are willing to adopt it of their own accord, we can then share the experience with industry, government, academia and research institutions, and scale up step by step.
As AI enters education, what is truly worth pursuing may not be “the fastest adoption” but “finding out as fast as possible what really works, and then letting what works spread naturally.”
- Original: Ray Lu’s personal Facebook post (published 2026-08-11)
Course titles follow the source. Check each course page for its teaching language.
Generative AI Leader Certification Learning Path
Five preparation courses are free and address business adoption and transformation. Course completion is not certification; the separate exam costs US$99 plus applicable tax. Its 90-minute duration is exam time, not course length.
AI capabilities and limitations
Understand next-token prediction, knowledge, working memory, and steerability to diagnose hallucinations and context limits. No technical prerequisites.