Should Children Learn to Think Before They Learn to Use AI?

Artificial intelligence is likely to be part of today's children's education, careers and everyday lives. But that does not automatically mean introducing it as early as possible is the right approach.

New York City has introduced a one-year restriction on student-facing generative AI for younger pupils during the 2026–27 school year, while older students move towards structured AI literacy and supervised use. That raises a more useful question than whether schools should simply ban AI:

The same question follows people into work. When does AI augment a skill, and when does it replace a skill that somebody never learned?

The short answer

What New York City is actually doing

The New York City Public Schools guidance is not a permanent blanket ban on artificial intelligence. It sets a one-year, age-sensitive policy for the 2026–27 school year and distinguishes between student-facing generative AI, approved educational software, assistive technology and the use of AI by staff.

For pupils in grades 2K through 8, software using student-facing generative AI is not permitted under the moratorium. NYCPS explains this as a way to protect human interaction, curiosity, hands-on learning and the effort through which children build foundational skills. The policy includes exceptions for assistive technology and other necessary tools for students with disabilities and English Language Learners.

Grades 9 through 12 take a different route. Student-facing AI is limited to approved, vetted programmes and guided use. Every high-school student is to complete two 45-minute AI-literacy modules covering how AI works, data privacy, bias and appropriate use in school. Career-readiness courses and approved pilots can provide deeper, supervised exposure.

Teachers and school staff can continue to use approved AI for instructional planning and operational work, subject to the school's privacy, security and procurement requirements. The guidance also draws a boundary around high-consequence decisions: AI cannot replace professional responsibility for grading, behaviour monitoring, placement, promotion, graduation or other decisions about students.

That is an important distinction. The policy is trying to separate early skill development from later AI-assisted learning. It is asking when AI belongs in the process, what the learner is meant to practise and who remains responsible for judging the result.

NYC Public Schools: Guidance on Artificial Intelligence (AI) and Screen Time →

This is not really an argument about banning technology

Schools have faced versions of this question before. Calculators changed how arithmetic could be done. Spellcheck changed how writing could be corrected. Search engines changed how information could be found. Sat-nav changed how journeys could be planned. None of these is an exact equivalent of generative AI, but each illustrates the recurring problem.

A tool can support learning, or it can remove the need to learn the underlying skill. A calculator is more useful after someone understands what an operation means. Spellcheck is more useful after someone has learned how sentences and words work. AI is more useful after someone has enough reasoning, research and subject knowledge to challenge what it produces.

AI can make good learners better

A cautious approach should not hide the genuine benefits of AI. Used well, it can help a student explore a difficult concept, receive an alternative explanation, generate examples, practise a language, test an idea or get rapid feedback on a draft.

It can help a curious learner move beyond the pace of a single textbook. It can make practice less intimidating by allowing someone to ask the same question in several different ways. It can also help a teacher prepare differentiated examples or identify where a class may need more explanation.

The useful word is support. AI should support thinking, not quietly replace it. A student who asks for three explanations and then compares them is doing something different from a student who submits the first answer without understanding it.

The danger is skill substitution

Generative AI can produce polished work before the user has developed the ability to produce or assess that work independently. That creates a gap between the appearance of competence and the underlying capability.

• Submitting an essay the student could not explain in a conversation.
• Producing code the developer cannot review, test or maintain.
• Generating research without judging the quality of its sources.
• Solving a maths problem without understanding the method.
• Producing an argument without understanding whether the evidence supports it.

In each case, the output may look better than the user's current ability. That is not always bad; drafts and examples can be useful. The risk appears when the output is mistaken for evidence that the person has learned the skill.

Knowing when AI is wrong is a skill

Generative AI does not know whether an answer is true simply because it can produce one. It can present incorrect information, invent details, follow weak reasoning, make confident mistakes, repeat a poor source or leave out an important qualification.

The user therefore needs enough understanding to challenge the output. That might mean knowing the subject, checking a primary source, testing the code, comparing independent evidence or asking a qualified person to review a high-consequence decision.

This is why prompting is not the whole skill. A well-written prompt can produce a more relevant answer, but it cannot guarantee accuracy. The user still has to understand what a good answer would look like and what evidence would change their mind.

AI literacy is more than prompt writing

Prompting matters. Clear context, a defined task and a request for assumptions can improve an AI response. But an organisation or school that treats prompt writing as the complete AI curriculum is teaching only the front door.

• What AI is good at and what it is poor at.
• Why a fluent answer can still be wrong.
• How to verify claims and judge source quality.
• How privacy, confidentiality and personal data can be affected.
• How bias and missing context can shape an output.
• When not to use AI at all.
• When human judgement, professional standards or safeguarding must take over.
• How much confidence the evidence actually supports.

That broader definition is more useful because it treats AI as part of a reasoning process rather than as a magic answer box. It also makes the lesson transferable: the same habits apply whether someone is checking an essay, a spreadsheet, a source list or a proposed business decision.

Why age and experience matter

This does not require unsupported claims about a particular age at which every person becomes ready for AI. The practical point is simpler. If someone has never developed a skill independently, there is a greater risk that AI becomes a substitute rather than an assistant.

A learner who already understands the basics can use AI to explore the edge of their knowledge. A learner who has not yet built the basics may have difficulty telling whether the explanation is sound, whether the method fits the problem or whether the answer merely sounds plausible.

So the useful question is not “AI: yes or no?” It is “AI: when, for what purpose and with what level of understanding?”

The same problem already exists at work

The school debate matters to businesses because people do not leave this problem behind when they enter employment. A junior employee can use AI to produce work that looks more experienced than they are. That can be helpful during learning, but it can also hide the gap until a decision, customer or system is affected.

Accountant

An experienced accountant using AI to analyse information can check the assumptions, reconcile the figures, identify missing context and explain the limits of the result. Someone relying on AI-generated financial analysis they do not understand is in a much weaker position, even if the report looks professional.

Developer

A developer using AI to accelerate coding can review the result, test edge cases, understand the dependencies and maintain the system later. Someone deploying code they cannot understand or test creates risk for security, reliability and future change.

Manager

A manager can use AI to explore options, challenge assumptions and summarise information. That is different from asking AI to make decisions the manager cannot independently evaluate. The tool can widen the conversation; it should not quietly become the accountable decision-maker.

Marketing

AI can accelerate drafting, research and ideation. Someone still needs enough knowledge to judge accuracy, evidence, tone, copyright risk and brand suitability. Fast production is not useful if the organisation publishes claims it cannot defend.

Are businesses creating AI dependence?

Most organisations should not respond by asking every employee to work without AI. The more useful response is to look for signs that the underlying capability is being lost or was never developed.

1. 1Are junior staff still learning the underlying task, or only learning which prompt produces an acceptable-looking output?
2. 2Could they complete the work if the AI tool disappeared for a day?
3. 3Can they identify a bad answer without waiting for somebody else to point it out?
4. 4Are experienced people reviewing high-consequence outputs?
5. 5Are employees learning judgement, or only prompting?
6. 6Are processes getting faster while capability quietly declines?

These questions do not assume that every employee should become an AI specialist. They ask whether the organisation retains enough knowledge to use AI safely, change tools when needed and recover when the tool is unavailable or wrong.

AI should change training, not remove it

Training should increasingly explain both the underlying process and the AI-assisted version of it. People need to know how the work is meant to happen, what good output looks like, where AI commonly fails and when an issue must be escalated.

• How the underlying process works.
• How to review AI output against evidence or a known standard.
• What a good result looks like before AI is introduced.
• Common failure modes, including invented facts and missing context.
• Verification and testing.
• Escalation when confidence is low or consequences are high.
• Appropriate data use and privacy boundaries.
• When human judgement is required.

Build enough understanding to use the tool intelligently. If the AI tool changes next year, the person should retain the skill and be able to learn the new interface rather than starting again from zero.

The simple IT Club test

The IT Club view

AI is going to be part of education and work. Trying to keep people away from it indefinitely makes little sense. Neither does assuming that earlier and more frequent AI use automatically creates better skills.

New York City's policy is interesting because it does not reduce the debate to technology versus no technology. It separates younger pupils who are still building core skills from older students who need structured AI literacy and controlled use. It also allows teachers to use approved tools while keeping professional responsibility with the people who understand the educational context.

For schools, that may mean protecting time for fundamental skills before introducing unrestricted AI assistance. For businesses, it means ensuring AI augments expertise rather than quietly replacing the expertise needed to judge its output.

Trying to work out where AI belongs in your business?

Ask the IT Club Advisor. One straightforward IT question. Independent guidance without the sales pitch.

Ask the Advisor →

Related reading

What Does Good AI Governance Look Like for a Small Business? →

How Can Businesses Reduce the Risk of AI Hallucinations? →

The Race to Replace: Why AI Needs to Start Doing the Work →

Sources and further reading

The policy details in this article were checked against the official NYC Public Schools guidance on 11 September 2026. The policy applies to the 2026–27 school year and may be updated as NYCPS evaluates its approach. This article is general technology information, not education, employment or legal advice.

NYC Public Schools: Guidance on Artificial Intelligence (AI) and Screen Time →

MKLINK: New York Bans AI For Most Schoolchildren For A Year →