Does Generative AI Use Stunt Cognitive Skill Development in Children?

We might still not know whether using generative artificial language models has any negative effect on cognitive skills, but Norway believes elementary school children should not be using it, and has banned it. 

Some studies suggest possible danger, but still inconclusive.

Study	Population	Key Finding	Relevance to Elementary Students
Bastani et al. (2025)	Nearly 1,000 high-school math students	Students using a standard GPT-4 tutor performed better while AI was available, but later performed 17% worse when AI access was removed. Researchers concluded that AI can become a "crutch" that impairs learning if poorly designed. (Scale)	Indirect evidence; suggests similar risks for younger learners who may rely heavily on AI assistance.
MIT Media Lab study (2025)	Adults (18–39) writing essays	AI users showed lower brain engagement, weaker memory of their own work, and reduced independent performance compared with non-AI users. Findings remain preliminary and are not specific to children. (MIT News)	Raises concerns that habitual AI use may reduce cognitive effort during learning tasks.
UNESCO (2023/2026 update)	Policy review	Warns that overreliance on generative AI may compromise the development of intellectual and social skills. Recommends age-appropriate use and safeguards. (UNESCO)	Directly discusses risks for children and recommends restrictions and human oversight.
Jaemarie Solyst et al. (2024)	26 middle-school girls	Participants initially showed substantial overtrust in ChatGPT outputs. Exposure to AI errors reduced that trust. (arXiv)	Suggests children may have difficulty evaluating AI-generated information critically.
Systematic review of generative AI in elementary education (2025)	Review of studies from 2020–2025	Found potential benefits but noted limited evidence, concerns about dependence, misinformation, and the need for teacher supervision. (Shanti Bhuana Journal)	One of the few reviews focused specifically on elementary education.

The emerging evidence points to three main concerns:

• Reduced productive struggle

• Learning often requires effortful practice, problem solving, and making mistakes.

• If AI supplies answers too quickly, students may skip the cognitive processes that build durable understanding. The high-school math study provides the strongest experimental evidence for this concern. (Scale)

• Cognitive offloading

• Researchers describe a phenomenon where people rely on external tools instead of developing internal knowledge and reasoning skills.

• Recent MIT findings suggest heavy AI assistance may reduce engagement and memory formation during learning activities. (MIT News)

• Overtrust and misinformation

• Children may be particularly vulnerable to accepting AI-generated content as authoritative.

• Studies of young users show that they can initially place excessive trust in chatbot outputs. (arXiv)

• 
  

But the evidence is not one-sided:

• Some studies find that AI tutors can substantially improve learning when designed to guide students through reasoning rather than simply provide answers. (Scale)

• The strongest "harm" findings generally occur when AI acts as an answer machine rather than a scaffold for thinking. (Scale)

• There is currently little direct experimental evidence involving elementary school children, so claims that generative AI definitely impairs learning in that age group remain tentative. Most experts argue that the impact depends heavily on how AI is designed and supervised. (UNESCO).

Current research does not show that generative AI inevitably harms elementary-school learning. 

However, several studies and policy reviews suggest that unsupervised or answer-focused AI use may impair skill development, critical thinking, and knowledge retention, particularly when students rely on it instead of engaging in the learning process themselves.

The strongest evidence so far comes from older students, while evidence specific to elementary-aged children remains limited and is still developing. (Scale).