Can AI Replace Teachers? Reflections on Teachers’ Day
How much teaching AI can replace deserves serious thought. Letting AI handle work it can perform reliably can give teachers more time for real problems, human development, and new educational possibilities. [中文]
Teachers’ Day is an occasion for gratitude and for reconsidering teachers’ work. If AI can explain clearly, answer questions whenever needed, organize practice, and provide feedback, we should ask what students still need teachers to do. An equally important question follows: why must teachers continue doing work that AI can already perform reliably? If technology creates an opportunity to leave substantial repetitive work behind, that opportunity deserves serious attention.
I understand the value of AI development in terms of human freedom: more time, greater capability, and more room to choose work worth doing. In education, this means handing over suitable tasks so that teachers can investigate problems, understand students, and organize practical experiences. Teachers also need time to learn and recover their energy. We should assess the change through both the education students receive and the space teachers gain to do good educational work.
Replacing a teacher therefore needs to be considered at several levels. Automating an explanation, changing a position, redistributing responsibilities within a school, and removing human participation from education are different propositions. AI may change or reduce some positions while making room for educational work that has long lacked time and resources. Students’ development should guide these decisions. Responsibilities and organizational arrangements can change as capabilities and circumstances change.
Repeated explanations, materials adapted to a learner’s progress, and timely feedback are important parts of good teaching. If AI can make them available to more students at lower cost, that is a real educational benefit. Technology may take over work that excellent teachers have invested considerable effort in doing well. Teachers need not prove their diligence by repeatedly performing activities that can be handed over. They can direct their attention toward worthwhile work they previously lacked the means to undertake.
Evidence already challenges the idea that AI must be an inferior substitute. A university physics trial published by Kestin and colleagues in 2025 found higher short-term learning gains in less time with a deliberately designed AI tutor across two lessons. This supports substitution for particular instructional tasks, without establishing replacement of an entire course or a lasting educational relationship. Original study
A high-school mathematics study adds a necessary qualification. Bastani and colleagues found that AI without learning safeguards could improve practice performance while harming later unaided performance; a design emphasizing hints mitigated the harm. What students can subsequently do for themselves matters alongside the answers they produce while supported. Original study
These findings are compatible. Different designs, learners, and tasks can produce different consequences. We need to compare specific educational arrangements, rather than an abstract human teacher with an abstract AI. An immediate test score cannot stand in for transfer, retention, sustained engagement, and independent judgment.
A frequently overlooked distinction explains why. An expert using AI at work may save effort while retaining the ability to judge the result. A learner who skips the same process may never have developed that judgment. Production asks how to obtain a result efficiently. Education must also ask what experiences enable a person to produce and judge results. These purposes are connected, but cannot simply be exchanged.
Independence does not require students to perform every task without tools. A reasonable aim is that they understand their purpose while using AI, can check consequential results, and know how to proceed when tools fail, advice conflicts, or conditions change. Education needs both foundational competence and judgment in using tools. Prohibiting a tool is not itself evidence that learning has occurred.
Imagine three students submitting equally functional programs. One can explain the design and repair a new failure. Another understands individual parts but not their relationships. The third has mainly copied generated output. Similar artifacts call for different educational responses. Explanations, modifications, transfer tasks, and continued observation can help a teacher identify what each learner needs. AI may assist these diagnoses and, in some circumstances, perform them better. Even understanding a learner cannot simply be declared an exclusively human ability.
To pursue the question fully, consider the strongest case for replacement. Suppose a future AI reliably identifies misconceptions, remembers a long learning history, gives appropriate encouragement, organizes collaboration, connects students with real projects, and demonstrably develops their independence. Where all these conditions are met, being a machine would not by itself supply a sufficient reason to reject substitution. We cannot place every important capability inside the category of nurturing people and then protect our conclusion by declaring that AI cannot do it.
Emotion needs similarly careful treatment. A human teacher may care deeply, but may also be exhausted, biased, or impatient. An AI response may be generated language and still help a student feel heard. Human identity does not guarantee better support. At the same time, some students value another actual person choosing to spend time with them, remembering shared experiences, and honoring commitments. Such a relationship has a significance that response speed does not capture. Whether and how to preserve it is a choice about educational values, not a permanent theorem about machine limitations.
The argument that AI cannot bear responsibility is not decisive for every position either. Schools, providers, and designated people can be assigned responsibility for an educational arrangement. Even where human accountability is required, it does not follow that a human teacher must perform every step of every lesson. The important questions concern who can hear an appeal, revise a mistaken assessment, and provide an effective remedy. Retaining a person who signs decisions but cannot correct them does not establish meaningful responsibility.
Many familiar arguments for irreplaceability therefore refer to different things: current unreliability, difficult physical circumstances, or a relationship that people wish to preserve. Each can justify human participation, but they are not the same kind of reason. A capability gap may narrow, the cost of handling a setting may change, and society can deliberate about the relationships it values. Stating these conditions clearly is more persuasive than asserting impossibility for all time.
Educational settings will not produce a single answer. An adult who can set goals independently and wants to learn a bounded skill faces a different situation from a young child learning to cooperate in a group. Laboratory safety, prolonged disengagement, and collective decisions involving conflicting values cannot be represented by a single exchange of questions and answers. Replacement happens in particular activities under particular conditions. Generalizing from one setting to all teachers, or using the hardest setting to defend every existing position, is equally unconvincing.
Teachers should therefore devote more attention to educational work that AI cannot currently perform reliably on its own and that still calls for actual human participation. This includes tasks beyond a system’s capabilities, as well as commitments, relationships, and collective decisions about educational purposes. We need to specify the work and allow its boundaries to change. Defending an ever-shrinking list of things machines cannot do leaves teachers reacting to each advance. Actively identifying unmet educational needs opens new possibilities for their work.
This brings us to the gap between teachers’ abilities and AI capabilities. Which difference should teachers seek to enlarge? Trying to remember more facts, generate more materials, or answer every question faster may demand enormous effort in the wrong direction. A more meaningful aim is to increase the additional growth made possible by a teacher exercising professional judgment and using AI appropriately, compared with what learners could obtain primarily through AI. Delegating suitable work and increasing a teacher’s contribution can reinforce each other: the time and capability gained can support deeper educational experiences.
For discussion, we can express this as a simplified relationship: a teacher’s educational contribution = outcomes with that teacher’s participation − outcomes without that teacher’s direct participation, with full access to AI. The comparison should involve similar learners, purposes, time, and resources. The alternative should be a well-designed feasible arrangement, not an intentionally weak chatbot. Teaching materials, institutional support, and basic safeguards remain in place; the question concerns this teacher’s direct contribution.
This comparison makes a deliberate choice. The teacher may use AI, so it measures a contribution in actual practice, rather than the unaided human brain against a model. Educational outcomes are not naturally reducible to one score either. Independent work, transfer, sustained engagement, and collaborative judgment all deserve attention. The figure simplifies them into one vertical dimension to illustrate a possible direction of development.
If a teacher’s practice and judgment remain static while AI support improves, an existing advantage may shrink or become a disadvantage. Another path is to use AI to gain time and expand capability, then use those conditions to design experiences that enable growth beyond what students achieve with tools alone. The widening gap represents that objective. Teachers have an opportunity to develop alongside technology without spending all their energy competing to retain the same repetitive task.
There is a counterintuitive consequence. When an excellent teacher’s methods are documented, shared, and incorporated into AI, they may raise the educational baseline and reduce that teacher’s relative advantage. This may be educational success. Teachers should not preserve a gap by withholding knowledge, manufacturing dependence, or obstructing better tools. They should seek new value, and be willing to take on different work when an earlier contribution becomes widely available.
The demand to keep enlarging the gap is therefore best understood as an invitation to renewal, not a guarantee that every position can survive indefinitely. Where an activity no longer requires human execution, reorganization should be possible. Conversely, a teacher may provide a small but important benefit to students in difficult circumstances. Its value does not disappear because it fails to produce a steeply rising curve. Decisions must also consider who benefits, whether the cost is reasonable, and who might be left behind by substitution.
Several directions seem worth exploring in a teacher’s own practice. They do not require outperforming AI in every capability. They ask teachers to turn the possibilities provided by tools into experiences through which students can develop.
The first direction is helping students discover problems worth pursuing. AI can propose topics and solutions. A teacher can take students into an actual setting to discover who encounters a difficulty, what existing explanations miss, and why the problem matters. A project to improve a campus service, for example, can begin with conversations with different users before a technical solution is chosen. A need omitted from a prompt may emerge through listening and observation. Teachers can connect knowledge with life and help students take responsibility for the problems they choose.
The second direction is designing learning that students need to experience themselves. AI can reduce the burden of finding materials and repeating explanations, but should not bypass the processes through which a learner develops judgment. Students might first propose an explanation and then compare alternatives from AI, reproduce a failure, adapt a design to a new constraint, or return later to a structurally similar problem. Teachers need to distinguish difficulties that merely consume effort from challenges through which capability grows. Reducing the former can make it easier to preserve the latter.
The third direction is making time for particular people. Time released from some explanations and grading can help a teacher notice the student who submits neat work but hesitates to speak, or the once-active student gradually withdrawing from collaboration. A careful conversation, an appropriate adjustment to a task, and continued follow-up may matter more than another set of exercises. AI can suggest observations and ways to begin a conversation. Generated words of concern do not themselves fulfill a commitment. A teacher’s actions can show a student that someone is willing to understand the difficulty and support another attempt.
The fourth direction is organizing real cooperation and shared life. Students need to listen to peers with different interests and views, negotiate responsibilities, honor commitments, and continue working through disagreement. AI can simulate perspectives and help prepare a discussion. Agreement with a simulation, however, is not agreement with the actual participants. Teachers can establish conditions for discussion, help students explain their reasons, protect voices that struggle to be heard, and support responsibility for a joint project. The classroom can become a place to learn how to live with other people.
The fifth direction is exploring unsettled questions together. A teacher can remain engaged with disciplinary developments and practice, or begin with a classroom observation, an unexpected result, or a student’s question. Teachers need not lead a research field to show how people approach what they do not know, use AI to check and compare ideas, and revise judgments in response to evidence. Participating in inquiry makes curiosity and respect for evidence visible in action.
These directions also need to be examined through their consequences. Away from the teacher, using another tool and facing a new situation, what can students still do? Did an intervention improve immediate performance while increasing dependence on hints? Students’ explanations, later performance, and differences between groups can guide gradual improvement. Teachers need not present every attempt as a success. Observing, adjusting, and trying again are part of professional growth.
A feasible beginning is to choose one repetitive, time-consuming activity in the next course, define the quality required, and let AI undertake a portion that the teacher can check. Reserve the time actually saved after checking and correction for an extended conversation, a real project, or the teacher’s own learning. After several weeks, reflect with students on whether the burden decreased, understanding deepened, and students became more willing to ask their own questions. Revise the arrangement if the savings are small; share effective practices with colleagues. Hope can begin with a modest change that actually works.
Teachers therefore need both professional thinking and deep thinking. Professional thinking supports diagnosis, task design, and assessment. Deep thinking asks why particular knowledge matters, what an assessment protects or neglects, and whether education develops autonomy or merely conformity. UNESCO’s AI competency framework for teachers includes human agency, ethics, pedagogy, and professional development. These express educational commitments, rather than proof of permanent human superiority. Framework overview
Such development needs institutional support. Schools should protect time saved through automation for education, learning, and rest, instead of immediately filling it with more forms, teaching hours, and formal deliverables. Introducing tools does not amount to liberation if the overall burden remains unchanged and teachers gain no discretion over their practice. Schools can reserve time for joint preparation, individual conversations, and small teaching experiments, allowing teachers reasonable room to try, fail, and revise.
These opportunities should reach more students as well. Students with fewer resources also need someone to follow their circumstances, not merely an automated service whose consequences nobody examines. If AI makes basic support more accessible, schools have greater reason to direct scarce human attention toward neglected needs while preserving ways to identify problems, challenge decisions, and obtain help. The distribution of the gains will shape whose freedom the technology actually expands.
Teachers’ authority will change too. As answers become more widely available, authority must depend increasingly on the reasons behind a judgment, the honoring of commitments, and the willingness to accept questions. A teacher can admit ignorance, use AI, or be corrected by a student. Authority is weakened when status prevents scrutiny or a student’s independent use of tools is treated as disloyalty. Teachers should seek to increase their educational contribution while reducing students’ uncritical dependence on a particular teacher or tool. The lasting value should belong to the student.
Whether AI can replace teachers deserves serious thought. So does the question of why teachers must continue doing work AI can perform reliably. We should welcome justified changes in how particular tasks are assigned, and assess changes to positions and educational relationships through learning, development, fairness, and accountability. Where a new arrangement is demonstrably better, preserving an old form alone is not a reason to reject it. Meanwhile, many worthwhile problems, practical experiences, and overlooked students await the capability teachers can release for them.
On Teachers’ Day, I hope teachers can see this possibility within the change: let AI do what it can do well, and undertake educational work that has yet to receive sufficient attention. A teacher finally has time to listen carefully to a question. A student ventures an original conjecture. A group learns to cooperate through a real project. These can be small sparks of hope. They may not immediately transform an education system, but they can change one classroom. As teachers keep learning and sharing, and schools protect room for exploration, the freedom technology offers can become something people experience in their lives. The value of teaching can continue to grow through these new practices.
Loading comments…
If comments are slow to load, view the discussion on GitHub.