When widely accessible generative AI arrived in classrooms, the first question many schools asked was:
How do we stop students from using it to cheat?
That was understandable.
A student could suddenly produce an essay, solve problems, summarize a reading, translate a passage, write code, or answer homework questions simply by describing the assignment to an AI system.
But education is now moving into a second stage.
Schools and universities are beginning to ask a harder question:
If students will live and work in a world where AI is widely available, what should they still learn without it—and what should they learn to do with it?
That question cannot be answered by banning every AI system.
It also cannot be answered by allowing AI to perform every difficult cognitive task for students.
The emerging consensus from organizations such as UNESCO and the OECD is more nuanced:
AI can support learning when used deliberately, but education must preserve human thinking, foundational knowledge, judgment, relationships, and agency.
The challenge is no longer simply whether AI belongs in education.
It is determining where it belongs, when it helps, and when it prevents the learning the assignment was supposed to produce.
The Short Answer
Generative AI is increasingly becoming part of education rather than something schools can treat as an external threat.
Students can use it for:
- tutoring;
- explanations;
- brainstorming;
- feedback;
- translation;
- study questions;
- coding help;
- research assistance.
Teachers can use it for:
- lesson planning;
- examples;
- differentiated materials;
- quizzes;
- administrative work;
- feedback drafts;
- professional learning.
But AI can also create serious problems involving:
- academic integrity;
- inaccurate information;
- overreliance;
- student privacy;
- bias;
- unequal access;
- weakened writing and reasoning skills.
The OECD's 2026 Digital Education Outlook concludes that generative AI can support learning when guided by clear pedagogical principles, but warns that it should not replace cognitive effort or the human relationships central to education.
| Question | Emerging answer |
|---|---|
| Should every student use AI? | Not necessarily |
| Should every use be prohibited? | Increasingly impractical |
| Can AI improve learning? | Yes, under some conditions |
| Can AI interfere with learning? | Yes |
| Is AI output always accurate? | No |
| Will teachers become unnecessary? | Evidence and policy guidance do not support that conclusion |
| Should students learn how AI works? | Increasingly yes |
| Does assessment need to change? | In many cases, yes |
The Calculator Analogy Helps—But Only a Little
AI is frequently compared with calculators.
When calculators became common, educators debated whether students would stop learning arithmetic.
Schools eventually developed a compromise.
Students still learn basic arithmetic.
They also learn to use calculators when the educational goal moves beyond manual computation.
That offers a useful principle:
Do not automate the skill while the student is supposed to be learning the skill.
A calculator can help a physics student focus on physics after the student understands basic arithmetic.
Likewise, generative AI might help a university student analyze alternative arguments after the student has learned how to construct an argument.
But AI is more disruptive than the calculator because it can operate directly on many of the skills education is designed to teach:
- writing;
- reasoning;
- explanation;
- research;
- coding;
- analysis.
Determining when AI should be used is therefore considerably harder.
The Central Question: What Is the Assignment For?
Suppose a student is assigned a 1,000-word essay.
The visible product is an essay.
But the real educational purpose may be to practice:
- reading;
- organizing evidence;
- constructing an argument;
- writing;
- revision;
- citation;
- critical thinking.
If an AI system performs those tasks, the student can submit a polished product while bypassing the learning.
The important question is therefore not:
Did AI help produce the essay?
It is:
Which cognitive work was the student expected to practice?
That distinction should guide AI rules.
AI Can Be a Powerful Tutor
The strongest argument for AI in education may be personalization.
A student who does not understand a concept can ask:
Explain this more simply.
Then:
Give me an example.
Then:
Don't tell me the answer. Ask me questions until I figure it out.
Then:
Show me where my reasoning went wrong.
A human teacher with thirty students cannot always provide unlimited one-on-one explanations.
An AI system potentially can.
The OECD's 2026 review identifies tutoring and personalized learning as among the promising educational uses of generative AI when tools are designed around learning rather than merely answer production.
That final condition matters enormously.
An Answer Machine Is Not Necessarily a Tutor
Suppose a student asks:
What is the answer to problem 7?
The AI gives it.
The student finishes faster.
Very little learning may have occurred.
Now suppose the system responds:
Show me how you started.
Then:
Which formula applies here?
Then:
Why do you think that step follows?
The technology is the same.
The pedagogy is different.
A genuinely educational AI should often make the student think more, not simply type less.
This is one reason the OECD distinguishes general-purpose generative AI from AI tools intentionally designed around learning science.
Writing Creates a Special Problem
Writing is both:
a communication product
and:
a way of thinking.
People often discover what they believe by trying to express it.
They notice contradictions.
They recognize missing evidence.
They restructure ideas.
If AI writes the first draft, some of that cognitive process may disappear.
But AI can also support writing without replacing it.
Students might:
- ask for feedback on an argument;
- compare two possible structures;
- request counterarguments;
- identify unclear sentences;
- practice revision;
- analyze weaknesses in AI-generated prose.
The educational objective should determine the rule.
Assessment Has to Change
For decades, schools often assumed that work completed outside the classroom represented the student's own unaided ability.
Generative AI weakens that assumption.
An essay submitted electronically may have been:
- fully human-written;
- lightly edited by AI;
- heavily rewritten by AI;
- generated from a prompt;
- generated and then extensively revised by the student.
This does not mean writing assignments should disappear.
It means assessment may need more evidence of the learning process.
Possible approaches include:
- in-class writing;
- oral defense;
- drafts;
- revision histories;
- student reflection;
- project-based work;
- personalized questions;
- demonstrations;
- discussion.
An instructor can ask:
Why did you make this argument?
A student who genuinely understands the work should be able to explain it.
AI Detection Is Not a Complete Solution
It is tempting to solve academic-integrity concerns using software that attempts to identify AI-generated text.
That approach has fundamental limitations.
AI writing changes.
Students edit output.
Models vary.
Human writing varies.
False accusations can carry serious consequences.
Schools therefore need policies based on:
- assignment design;
- clear expectations;
- evidence of learning;
- conversations with students;
rather than assuming a detection score can definitively determine authorship.
Schools Are Moving From Prohibition Toward Governance
The institutional response has evolved quickly.
In the early generative-AI period, many institutions focused primarily on restrictions.
By 2025 UNESCO reported that nearly two-thirds of surveyed higher-education institutions in its Chair and UNITWIN networks either had formal AI guidance or were developing it.
That does not mean every school has embraced AI.
It means institutions increasingly recognize that they need explicit rules.
Good policies answer questions such as:
- When may students use AI?
- Must AI use be disclosed?
- What counts as unauthorized assistance?
- What student information can be entered?
- Which tools are institutionally approved?
- How should AI-assisted work be cited?
- What remains the student's responsibility?
Ambiguity is often worse than either permission or prohibition.
The U.S. Department of Education Has Shifted Toward Responsible Use
In July 2025, the U.S. Department of Education issued guidance explaining that existing federal education funding can support appropriate AI uses when those uses comply with applicable legal requirements.
The guidance identified potential applications across instruction, tutoring, advising and educational operations while emphasizing responsible implementation.
That reflects a broader change.
The policy question is increasingly:
How should education use AI responsibly?
rather than:
How can education keep AI outside the building?
AI Literacy Is Becoming Part of Literacy
Students need more than the ability to type prompts.
UNESCO's AI Competency Framework for Students identifies twelve competencies across areas designed to help students become responsible users and co-creators of AI.
Useful AI literacy includes understanding:
- what AI can do;
- what it cannot reliably do;
- why answers can be wrong;
- bias;
- privacy;
- intellectual property;
- verification;
- appropriate disclosure;
- human responsibility.
A student who trusts every AI answer is not AI-literate.
Neither is a student who knows how to generate polished content quickly but cannot evaluate it.
Teachers Need AI Literacy Too
Teachers cannot effectively guide students through technology they do not understand.
UNESCO's teacher framework organizes AI competence across five dimensions:
- human-centered mindset;
- ethics;
- AI foundations and applications;
- AI pedagogy;
- professional learning.
Notice what is absent:
prompt tricks.
The emphasis is broader.
Teachers need to know when AI improves pedagogy and when it undermines it.
AI Can Reduce Teacher Workload
Teaching contains large amounts of preparation and administration.
AI may help teachers:
- draft lesson outlines;
- generate examples;
- create differentiated reading levels;
- prepare practice questions;
- produce rubric drafts;
- summarize material;
- generate parent-communication drafts;
- brainstorm activities.
Used carefully, this can give teachers more time for the distinctly human parts of teaching:
- conversation;
- explanation;
- encouragement;
- observation;
- judgment;
- mentoring.
The objective should be to automate low-value workload, not human relationships.
Teachers Remain More Than Information Sources
A student can ask AI:
What caused World War I?
That does not make the history teacher unnecessary.
A teacher does more than transmit information.
Teachers:
- recognize confusion;
- motivate reluctant learners;
- manage groups;
- understand local context;
- create expectations;
- resolve conflicts;
- notice emotional changes;
- determine what should be taught next;
- model intellectual behavior.
Education is social.
The OECD explicitly emphasizes preserving human relationships at the center of learning even as AI becomes more capable.
The Hallucination Problem Matters More in Education
Generative models can produce inaccurate information confidently.
Experienced professionals may recognize errors.
Students often cannot.
That creates a dangerous asymmetry.
The learner uses AI precisely because the learner does not know the subject well enough to answer independently.
But that also means the learner may not know when the AI is wrong.
Education therefore needs to teach:
verification as a core AI skill.
Students should learn to ask:
- What is the source?
- Can I verify this elsewhere?
- Is the citation real?
- Does the quotation actually appear in the source?
- Does the answer contradict the textbook or primary material?
- What assumptions is the system making?
Privacy Matters
Students can unintentionally submit sensitive information to AI tools.
Examples include:
- names;
- grades;
- disabilities;
- behavioral information;
- personal writing;
- health information;
- identifiable classroom records.
Schools must therefore evaluate AI not merely as an educational tool but as a data system.
Appropriate use depends on privacy rules, tool contracts, institutional policy and student age.
“Just paste it into the chatbot” is not an acceptable universal workflow.
Equity Can Improve—or Worsen
AI may expand access to:
- tutoring;
- translation;
- accessibility support;
- writing assistance;
- specialized explanations.
That could be especially valuable where individualized educational support is scarce.
But unequal access can create a different outcome.
Students with:
- better devices;
- paid AI subscriptions;
- faster internet;
- more AI-literate parents;
- stronger prior knowledge;
may benefit more than students without those advantages.
UNESCO consistently frames AI education policy around inclusion and human-centered access precisely because technology alone does not guarantee educational equity.
Students Still Need Knowledge in an AI World
A common argument says:
Why memorize anything if AI can tell you the answer?
Because knowledge is part of thinking.
A person cannot evaluate an AI explanation of biology without knowing biology.
They cannot identify a fabricated historical claim without historical knowledge.
They cannot judge incorrect code without understanding programming.
They cannot notice a bad argument without knowing how reasoning works.
External tools can extend knowledge.
They do not eliminate the need for internal understanding.
The OECD's 2026 guidance emphasizes maintaining foundational knowledge and independent thinking even while teaching students to work with AI.
A Useful Three-Stage Learning Model
One practical approach is:
1. Learn without AI
Students first develop the foundational skill themselves.
2. Learn with educational AI
AI supports practice through tutoring, feedback and questioning.
3. Work with general-purpose AI
Once the student understands the domain, AI becomes a productivity and exploration tool.
The OECD's 2026 Digital Education Outlook explicitly proposes a similar progression: develop valued human skills without GenAI, with educational GenAI, and then with general-purpose GenAI.
This provides a useful middle ground between prohibition and unrestricted use.
Five Misconceptions About AI in Education
“AI is just cheating software.”
It can facilitate cheating, but it can also provide tutoring, feedback, translation and accessibility support.
“Students no longer need to learn writing because AI can write.”
Writing is part of learning to organize and evaluate thought.
“AI tutors will replace teachers.”
A tutoring system can provide explanations. Teaching includes relationships, motivation, judgment and social context.
“The solution is perfect AI detection.”
Detection cannot serve as a universal proof of authorship.
“Schools should either ban AI or allow everything.”
Different learning objectives justify different rules.
What Students Should Learn
A useful AI curriculum should teach students to:
Ask good questions
Clear requests improve interaction.
Verify answers
AI fluency without skepticism is dangerous.
Understand uncertainty
A confident answer can still be wrong.
Protect privacy
Students need to know what information should not be uploaded.
Disclose appropriately
Academic work should follow the institution's rules for AI assistance.
Preserve their own thinking
AI should extend understanding rather than substitute for it.
What Teachers Can Do
State the AI rule for every major assignment
Use categories such as:
AI prohibited
AI allowed for brainstorming
AI allowed with disclosure
AI encouraged as part of the task
Students should not have to guess.
Ask students to show process
Drafts and explanations reveal learning better than polished final products alone.
Use AI critically in class
Give students an AI answer and ask them to find its errors.
Design assignments around local context
Personalized and experiential tasks are harder to outsource meaningfully.
Preserve some AI-free practice
Students still need opportunities to build unaided skills.
What Schools and Universities Need
Institution-level governance should address:
- approved tools;
- student privacy;
- academic integrity;
- accessibility;
- age restrictions;
- data retention;
- teacher training;
- assessment;
- disclosure;
- procurement;
- bias;
- security.
AI policy cannot remain entirely the responsibility of individual teachers.
What to Watch Next
AI-native tutoring systems
Will systems designed around pedagogy outperform general-purpose chatbots for learning?
Assessment redesign
Schools may increasingly evaluate process, discussion and demonstration alongside submitted products.
AI competency standards
AI literacy is likely to become part of formal curricula.
Personalized education
AI could offer more individualized pacing and feedback, but effectiveness needs rigorous evaluation.
Teacher workload
One of AI's clearest near-term benefits may come from reducing repetitive preparation and administration.
Cognitive dependence
Researchers will need to study whether intensive AI use weakens independent writing, reasoning and problem-solving in some contexts.
The Bottom Line
Generative AI has made one thing clear:
Education cannot evaluate learning only by looking at the finished product.
A polished essay does not necessarily prove the student can write.
Correct code does not necessarily prove the student can program.
A good answer does not necessarily prove understanding.
That sounds like a threat.
It can also be an opportunity.
Schools can refocus on what education was supposed to produce in the first place:
knowledge, reasoning, curiosity, judgment, communication and the ability to learn.
AI can support those goals.
It can also bypass them.
The difference depends less on whether AI is present than on how the learning experience is designed.
The most useful educational question is therefore not:
Should students use AI?
It is:
At this stage of learning, what thinking must the student still do for themselves?
Questions People Ask
Should students be allowed to use AI for homework?
It depends on the learning objective and teacher policy. AI may be appropriate for feedback or tutoring while being inappropriate when the assignment is specifically testing unaided writing or problem-solving.
Can AI improve learning?
Research reviewed by the OECD suggests it can when used according to clear pedagogical principles, particularly when AI supports rather than replaces cognitive effort.
Will AI replace teachers?
Current educational guidance instead treats AI as a potential support tool while emphasizing the central role of human teachers and relationships.
How are universities responding?
Policies have expanded substantially. UNESCO reported in 2025 that nearly two-thirds of surveyed institutions in its higher-education networks had AI guidance or were developing it.
What is AI literacy?
It includes understanding AI capabilities and limitations, verifying outputs, recognizing ethical and privacy issues, and knowing how to use AI responsibly rather than merely knowing how to prompt it.