Gone With the Bot: Ban the Wind or Teach Kids how to Fly?

This week the New York City Department of Education did what regulators tend to do when a technology outpaces them: it drew a line in the sand. The city's new AI policy bans student-facing generative AI tools- chatbots, AI tutors, the whole lot- for every child from 2-K through eighth grade, and heavily restricts high schoolers, with a few pilot exceptions. Individual devices are barred for pre-K through second grade. Screen time on personal devices is capped at 30 minutes in grades 3–5 and 45 minutes in middle school (Chalkbeat).

I've been teaching since the technological-abucus days (back when the internet didn't even have advertisements) and I've seen this movie before. The names change (television, video games, smartphones, social media, now ChatGPT), but the plot beats are the same: a technology arrives, adults panic, policy chases the panic, and kids keep using the thing anyway.

So let me say the obvious thing out loud.

The weather report already said it would be windy. The kids already bought the kites. Trying to ban the wind is a great way to look decisive and a terrible way to actually change the weather.

What the Policy Actually Says

Give the DOE some credit - this isn't a chaotic edict. The framework has real structure:

• No student-facing generative AI for 2-K through 8th grade, with exceptions for state reading and math assessments, English language learners, and students with disabilities.

• High schoolers get restricted access, with permitted use in literacy, career-readiness programs, and five pilot schools.

• Teachers can use approved AI tools for lesson planning, translation, and drafting communications- but not for grading, behavior monitoring, counseling, or writing special-education plans.

• An implementation task force is being convened to figure out the rest (Chalkbeat).

Chancellor Kamar Samuels, to his credit, admitted in May that the earlier guidance had "missed the mark." More than half of the City Council signed a letter urging Mayor Mamdani and the chancellor to pause AI in schools, and a preliminary policy in March drew ferocious backlash from parents and grassroots groups (Chalkbeat). So the political pressure was real, and the instinct to protect kids from a technology whose long-term effects we barely understand is fundamentally a caring one.

Here's the problem: caring isn't the same as effective.

Who Are We Actually Trying to Keep Up With?

Ask yourself who this policy is really racing against. The kids? The technology? Other countries? Compliance officers with clipboards?

Because the kids are already gone. Nearly nine in ten children ages 9–17 use generative AI, and a quarter use it daily, according to Common Sense Media's 2026 census (Fox Baltimore). The College Board found 84% of high schoolers use it for schoolwork, up from 79% in a matter of months (College Board). Pew clocked 54% of teens using chatbots for schoolwork, with one in ten doing "all or most" of their homework that way (Pew Research Center). OpenAI's own analysis put U.S. student ChatGPT prompts at more than 460 million messages per week during the school year, with a distinct Sunday-night spike (Tun.com).

Meanwhile, teacher use of generative AI jumped from 6% of schools in 2022 to 90% by 2025 (University of Chicago Becker Friedman Institute).

A ban doesn't stop the technology. It just moves it off school grounds, off the record, and out of the reach of the adults who could actually teach kids how to use it well.

The Case for the Policy (It's Not Nothing)

The DOE isn't wrong to be cautious. The research the skeptics keep citing is real- and it should concern all of us.

A landmark 30-month study of 26,811 Chinese students in grades 7–12 found that AI use boosted homework scores by 18% and cut homework time by 30%- but closed-book exam scores dropped 20% within six months, and high-stakes entrance exam scores fell 18% to 24% after about two years (CEPR; World Bank).

A Wharton-led experiment cited by the OECD found students using GPT-4 improved short-term performance by up to 48% - but scored 17% worse than peers once the AI was taken away (European Parliament). A randomized trial called it exactly what it looks like: a cognitive crutch (SSRN).

The DOE's presentation put it plainly: "Younger students need hands-on learning, social interaction, play, and teacher-guided instruction. Routine individual screen use can crowd out those experiences" (Chalkbeat). They're not wrong about that. Six-year-olds do not need a chatbot. They need blocks, books, and a grown-up who asks good questions and knows how to genuinely listen so they feel understood and important.

So if the policy said, "Let's protect the youngest learners from AI outsourcing while we build something smarter for older kids" - I'd be nodding.

That's not quite what it says.

The Case Against (Or, the Screen Time Déjà Vu)

Here's where a veteran educator gets to be a little skeptical.

As a child, television was the villain; a generation ago, adults vilified computer screen time. Then it was video games. Then screen time. Then phones. The adult response: limit it. Every time, the initial policy instinct was restriction by the clock- two hours here, no screens there.

And every time, the research eventually caught up and said: it was never really about the minutes. The American Academy of Pediatrics moved away from rigid time limits in 2016 in favor of a quality-and-context framework, because the evidence showed that what children do on screens and who they do it with matter far more than how long (American Academy of Pediatrics). Passive scrolling harms attention and language; interactive, co-viewed, purposeful screen use is neutral to positive (Children and Screens; Canadian Paediatric Society summary).

A 30-minute daily cap for third graders sounds tidy on a policy memo. In practice, it treats a coding lesson and an autoplay YouTube binge as equivalent. They are not.

The Wharton and Chinese studies also carry a nuance the headline version misses: the damage clustered among students who used AI to outsource thinking- not among students who used it to explain, brainstorm, or check (CEPR; arXiv). One arXiv field study put it flatly: LLMs increased learning when students asked for explanations, and decreased it when they asked for solutions (arXiv).

That is not an argument for a ban. That's an argument for teaching kids how to use it.

Which most schools aren't. Only 30% of teen AI users say a teacher has ever discussed with them how to use AI safely (WINS Solutions summary of Common Sense). We're banning the tool without teaching the tool. That's the education-policy equivalent of banning the pool without teaching swimming.

Go Slow to Go Fast: A More Grown-Up Path

Here's the good news for busy leaders and new teachers alike: you don't have to pick between "let the robots teach the kids" and "outlaw the kites." There's a middle lane. Call it intentional embedding.

1. Start with the tasks that already require thinking. Embed AI into work that demands critical thinking, collaboration, or genuine inquiry- a Socratic seminar prep, a lab-notebook analysis, a debate rebuttal. AI as a sparring partner, not a ghostwriter. This is the "explanation, not solution" pattern the research keeps rewarding (arXiv).

2. Protect the productive struggle. The World Bank calls this "desirable difficulty" for a reason- the moments kids feel stuck are precisely where learning lives (World Bank). Design assignments where AI cannot short-circuit the struggle: oral defenses, in-class writing, live problem-solving. Make the thinking visible.

3. Teach AI literacy the way we teach source evaluation. The DOE's own presentation admits high schoolers "need guided preparation for a world and economy increasingly driven by AI…paired with instruction on how AI works, its biases and risks, and how to evaluate its outputs" (Chalkbeat). Good. Now do that in eighth grade. And seventh. And sixth.

4. Invite the actual stakeholders in. Here's a radical idea (gasp): let educators and parents help write the rules. The people using AI at home with their kids and the teachers watching it land in the classroom have more field data than any task force convened in a conference room. Some of NYC's fiercest AI critics are parents who are AI professionals. Their expertise is a gift, not a threat.

5. Distinguish the tool from the crutch. A calculator didn't kill math class- it changed what math class was for. AI can be a calculator for language and analysis, if we teach kids to check its work, question its answers, and never let it do the part of the task where their brain was supposed to grow.

What Nothing- Repeat, Nothing- Replaces

I'll say the quiet part loud, because after working with students since the 1900s, I've earned it: no policy, no ban, no permission slip, no shiny AI tutor beats a great teacher.

Active engagement. Collaboration. Productive struggle. A grown-up who notices when a kid's eyes glaze over and adjusts on the fly. That is the actual technology of learning. AI is, at best, a very capable assistant to that work. At worst, it's a distraction dressed as a shortcut.

Be intentional about how and when. Use AI to translate a family letter into eight languages in eight seconds so a parent can actually read it. Don't use it to "differentiate" your way out of knowing your students. Use it to draft the first version of a rubric you then rewrite with your team. Don't use it to write feedback to a child whose face you should be looking at.

Your Move

So here's where I hand the kite string back to you.

If you're a leader: Don't just implement the policy - interpret it. Build a school-level AI literacy plan. Convene your teachers and parents. Pilot the intentional-embedding approach in one grade, measure what happens, and scale what works. The DOE is forming a task force; put your hand up.

If you're new to your career: You have a superpower the veterans envy. You grew up with this stuff. Use it. Learn AI literacy the way older teachers learned pedagogy- deliberately, curiously, with your students in mind. When your school asks who wants to co-design an AI-integration pilot, be the one who says yes.

If you're a parent: The most powerful AI policy in your child's life is the one you set at your kitchen table. Ask what they used it for today. Read the output together. Correct it together. Teach them that "the machine said so" is not an answer.

The wind is here. The kites are up. The question isn't whether we let the wind blow. It's whether we teach our kids how to fly.

References

Common Sense Media and SSRS. "The Common Sense Media Census: AI Use by Tweens and Teens (2026)." June 8, 2026. https://ssrs.com/news/the-common-sense-media-census-ai-use-by-tweens-and-teens-2026/.

College Board. "New Research: Majority of High School Students Use Generative AI for Schoolwork." October 6, 2025. https://newsroom.collegeboard.org/new-research-majority-high-school-students-use-generative-ai-schoolwork.

European Parliament. Artificial Intelligence in Classrooms: Cognitive Dimensions. 2026. https://www.europarl.europa.eu/cmsdata/304836/AI%20in%20Classrooms_Cognitive_Dimensions_final.pdf.

Ju, Songqi, and Ying Bai. "The Generative AI Learning Penalty." CEPR Discussion Paper DP21577. June 2, 2026. https://cepr.org/publications/dp21577.

Lehmann, Matthias, Philipp B. Cornelius, and Fabian J. Sting. "How Students' Use of ChatGPT Affects Their Learning." arXiv preprint, 2024. https://arxiv.org/html/2409.09047v1.

Nagata, Anaya M. "A Warning Shot for Human Capital: Evidence of an AI Learning Penalty." World Bank Blogs, June 18, 2026. https://blogs.worldbank.org/en/investinpeople/a-warning-shot-for-human-capital--evidence-of-an-ai-learning-pen.

Nasr, Nour. "ChatGPT as a Cognitive Crutch: Evidence from a Randomized Controlled Trial on Knowledge Retention." SSRN Working Paper, July 21, 2025. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5353041.

Pew Research Center. "How Teens Use and View AI." February 24, 2026. https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/.

American Academy of Pediatrics. "Screen Time Guidelines." Center of Excellence on Social Media and Youth Mental Health. Accessed September 2, 2026. https://www.aap.org/en/patient-care/media-and-children/center-of-excellence-on-social-media-and-youth-mental-health/qa-portal/qa-portal-library/qa-portal-library-questions/screen-time-guidelines/.

Children and Screens. "Are Some Types of Screen Time Better Than Others?" February 23, 2024. https://www.childrenandscreens.org/learn-explore/research/are-some-types-of-screen-time-better-than-others/.

Ponti, Michelle. "Quality of Kids' Screen Time Matters as Much as Quantity." CMAJ 191, no. 25 (June 24, 2019). https://pmc.ncbi.nlm.nih.gov/articles/PMC6592813/.

Rugaber, Christopher. "OpenAI Report: 460M Weekly Student Messages Sent to ChatGPT." The University Network, August 27, 2026. https://www.tun.com/home/openai-report-460m-weekly-student-messages-sent-to-chatgpt/.

Sanchez, Iris, and Aloha McBride. "AI Diffusion Gaps: Unequal Integration of AI Across K-12 Schools." Becker Friedman Institute, University of Chicago, August 5, 2026. https://bfi.uchicago.edu/insights/ai-diffusion-gaps-unequal-integration-of-ai-across-k-12-schools/.

Winssolutions. "Survey: 70% of US Teens Use AI for Schoolwork." August 19, 2026. https://www.winssolutions.org/teens-using-ai-for-schoolwork-survey/.

World Bank. "Is AI Making Us Smarter or Just Making Us Look Smart?" November 20, 2025. https://blogs.worldbank.org/en/education/is-ai-making-us-smarter-or-just-making-us-look-smart-.

Zimmer, Amy, and Alex Zimmerman. "NYC to Ban Student AI Tools in 2-K through 8th Grade and Set Screen Time Limits." Chalkbeat New York, September 2, 2026. https://www.chalkbeat.org/newyork/2026/09/02/nyc-schools-to-set-ai-policy-ban-screen-time-limits/.