Students use AI to complete homework while teachers use AI to grade it, ultimately competing to see whose AI is better and more advanced. This may seem like a joke, but in fact, since the birth of large AI models, how to distinguish the AI generation

According to media reports, a batch of detection tools represented by GPTZero are rapidly rising, and they are no longer simply giving AI generation probability, but can accurately identify the generation trajectory of text. Various detection tools a

Generative AI is pushing us down a steeper and broader path of education change. For educators and educatees, what needs to be changed is not only the way of teaching and learning, but also the whole idea of education.

The attack-defense upgrade between teachers and students has not ended

In the tug-of-war between teachers and students around teaching, AI and human teachers are driving a continuous iteration of offensive and defensive upgrades, which is even referred to as an AI arms race between teachers and students.

On the student side, AI is evolving from "direct generation" to "multi-step processing": it is first used to retrieve textbooks and authoritative sources, then use hint engineering to induce organizational reasoning, and then adjust personality corpu

On the teacher's side, they can detect the shift from a single 'AI rate' to an 'evidence puzzle'. Teachers can also utilize AI to combine multi-modal signals such as confusion level, syntactic tree diversity, reference stability, citation searchabili

Even further, teachers will shift to "process-oriented" teaching design in teaching. Many teachers will require students to submit thought chains, draft versions and oral retelling, and incorporate AI into the controllable process of "generation-veri

In this situation, some platforms that once attempted to offer gray services by providing AI-powered humanized tools are finding it difficult to continue doing so.

Not banning students from using AI, but encouraging them to be responsible for their use

When students can use AI to complete tasks that require their own thinking, and when AI can not only set questions but also solve them, the evaluation of students' learning outcomes through traditional objective questions, programming questions, and

However, we should not simply attribute this to an upgrade of cheating tools. Instead, we should accept that this is a structural redistribution of problem-solving power: in an era where answers are easily accessible, the value of learning lies not i

At this point, if our evaluations of student learning and teacher performance are still results-oriented, the AI game between teachers and students will quickly evolve into a competition of who calls for it faster. Only when we shift our evaluations

This means that in the future teaching process, there will be more open tasks, project-based assignments, oral defenses, and on-the-spot evaluations. It also means that when submitting assignments, students need to attach 'AI intervention explanation

The 'hidden value' of good teachers will become more evident

In fact, AI is reshaping educational delivery in many schools.

For students, AI can only be a learning assistant; for teachers, AI can only be a co-pilot. AI can take over large-scale, standardized creative tasks, allowing teachers to spend their time on more professionally valuable areas. For example, building

In the past, the 'hidden value' of good teachers, which could not be replicated on a large scale, including their professional judgment and humanistic care, has been placed in a more prominent position. Taking the example of the TeachMaster intellige

Reconstructing the Roles of 'Passing on Knowledge, Teaching Skills, and Solving Doubts' with Human-machine Collaboration

Although the promotion of AI in teaching and learning can make high-quality educational resources more widely available, there are still four challenges in the implementation of generative AI in teaching and learning processes:

One is the knowledge granular gap. Currently, most general models are strong in breadth rather than depth. To introduce AI tools into high-level specialized teaching, it is necessary to compensate through domain knowledge bases, expert labeling, and

The second is the crisis of evaluation system. When AI achieves "instant solution", the validity of standardized testing decreases. Therefore, we need to take oral explanation, open-book examination, limited resources, and traceable evidence chain as

Thirdly, policies and governance are not yet perfect. Data privacy, copyright ownership, and responsibility demarcation all require clear institutional safeguards. Trusted model white lists, hierarchical authorization, and log auditing should be esta

Fourth, there is algorithmic aversion and organizational inertia. Often, institutions or the public resist not just technical issues, but a combination of values, systems, and fields.

In the rapidly advancing era of AI, education should collaborate between humans and machines to reshape the division of labor in teaching morality, imparting knowledge, and solving doubts. Teaching morality means that teachers grasp the value orienta

Technology does not lead education to be dehumanized. On the contrary, it brings people's work back to human beings themselves, including judgment, empathy, encouragement, and the persistence of values.

Currently, China is moving from being a big educational country to being a strong one. When concepts and technologies align, supply and evaluation move forward together, and teachers collaborate with AI, education supply will undergo a transformation

（Author is an associate professor at the School of Artificial Intelligence, Shanghai Jiao Tong University）