Does AI Help Children Learn, or Harm Them? It Depends on Their Age, a New Harvard Study Says

Does AI Help Children Learn, or Harm Them? It Depends on Their Age, a New Harvard Study Says

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
16. 11. 2026
5 minutes reading · 6 views
Does AI Help Children Learn, or Harm Them? It Depends on Their Age, a New Harvard Study Says

Harvard’s Graduate School of Education has published a guide intended to help parents and teachers decide when artificial intelligence helps children learn and when it may do more harm than good. It is called AI Guidance by Age and is based on more than five thousand academic papers covering 81 areas of child development. However, the authors readily acknowledge that these are not definitive findings but hypotheses, because thorough research does not yet exist.

That is precisely why the guide was created. According to the authors, obtaining reliable evidence about what AI does to the developing mind may take ten years or more. We have already experienced a similar situation, as an entire generation grew up with social media before long-term studies became available. Yet the decisions parents and schools make today will be difficult to reverse. Project lead Jeff Curley summed it up by saying that an academic institution cannot wait, because teachers and parents are looking for answers right now.

One question instead of a list of prohibitions

For the average parent, the most useful takeaway from the entire guide is a simple rule of thumb proposed by co-author Tessa Forshaw. Instead of drawing up a list of what is and is not allowed, she offers a single question.

Does this use of AI lead the child toward practicing the skill and engaging with the people around them, or away from them?

If it leads toward them, she says such use is generally fine. If it leads away from them, it is worth thinking about it again. This is based on two principles that Forshaw says apply to learning regardless of technology. Children need close contact with more experienced people, and they need opportunities to actually practice a skill.

This question works even if you have no idea how a chatbot operates. If a child asks it to explain the process and then tries it independently, they are practicing the skill. If they have it complete the assignment and copy the result, they have skipped the practice.

What is being developed at each age

The rest of the guide is based on the idea that children develop different things at different ages, so the same technology may help at one stage and cause harm at another. The guide therefore combines two dimensions: the child’s age and the ability they are currently developing, from early literacy to emotional regulation. For each combination, it answers four questions: What is developing at this age, what does existing AI research say about it, how might AI disrupt it, and how might AI help?

This can be illustrated with children between the ages of nine and eleven. At this age, abilities such as working memory, sustained attention, and monitoring one’s own thinking are still developing. According to the guide, AI can be beneficial here when adaptive programs keep the difficulty level exactly at the point where the child is challenged but can still manage the task. It also helps when the child evaluates whether an AI-generated answer is correct, because this allows them to practice critical thinking.

The situation is reversed with homework tools. According to the authors, these tools short-circuit memory retrieval—the very uncomfortable moment when a child has to recall something independently. Yet it is precisely this moment that builds memory and leads to learning. The child receives a finished result but misses the part for which the assignment was created in the first place.

How the guide was created and why caution is needed

The authors described their process quite openly, probably in part because otherwise someone might accuse them of writing about AI with the help of AI. They adopted the principle of putting AI into the human loop, rather than putting humans into the artificial one.

They first compiled a library of more than five thousand academic papers from developmental psychology, cognitive science, and neuroscience, supplemented by early research on AI in education. AI summarized what was known from this library about each topic and age group, and the result was given to a research assistant, who wrote the text. A second assistant reviewed it, an editor shortened it from five pages to two and converted it into accessible language. AI then checked every citation against its source and flagged any discrepancies. Finally, everything was reviewed by someone with a doctorate or a faculty member.

Another interesting step was one that few people would expect in an academic project. The authors brought in a coach whose job was to make them stop clinging to their initial ideas and to the assumption that everything would remain unchanged. They wanted this to broaden their thinking about the many ways AI might help or harm development.

The guide also has its limitations, which the authors themselves acknowledge. It focuses primarily on neurotypical children and young people under the age of forty, and most existing developmental research comes from countries in the Global North, so the findings may not apply to other settings. Moreover, each entry is a hypothesis—the most likely explanation assembled from incomplete evidence—not a verified conclusion.

Reactions among teachers have so far been mixed. Special education teacher Heather Gauck from Michigan described the guide as a useful deep dive into the developmental stages essential for learning. Tim Wells, an assistant principal at a high school in Nevada, wrote that although the concept is interesting, the tool seems fairly basic and he would not recommend it yet, because ChatGPT serves him better for the same purpose.

This only reinforces the uncertainty surrounding the research. A review by Stanford’s center for AI in education showed that existing findings are mixed and depend heavily on how a tool is designed and used. A separate review of meta-analyses on the impact of AI on classroom learning found that the existing studies contain errors and cannot serve as a basis for decision-making. The Harvard team therefore describes the guide as a living resource that will be updated as new studies and feedback from teachers become available.

Sources: harvard.edu, timesofindia.indiatimes.com and edweek.org

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

OpenAI introduces GPT-6.2 Astra and Sol for professional and everyday workOpenAI introduces GPT-6.2 Astra and Sol for professional and everyday work
OpenAI has introduced two models: Astra for complex professional tasks and Sol for everyday work. Both are available in ChatGPT Work, Codex and through the API.
3 min read
17. 11. 2026
Tencent releases Hy4 Flash with image support and a context window of up to 1 million tokensTencent releases Hy4 Flash with image support and a context window of up to 1 million tokens
The lighter Hy4 Flash model targets chat, documents and routine code edits. It accepts text and images, supports tool use and, according to Tencent, matched Claude Haiku 5.5 in tests of everyday tasks.
1 min read
16. 11. 2026
Companies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia ChipsCompanies Linked to Chinese Spy Invested in DeepSeek and Promised Banned Nvidia Chips
Companies linked to an alleged Chinese spy joined DeepSeek’s latest funding round. In exchange for a stake, they reportedly promised banned Nvidia Blackwell chips.
3 min read
16. 11. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok