# What Science Says Since the release of LLM models in 2023, scientists have been studying how it affects learners. ![Many children sit on a carpet in a classroom facing the teacher at the front of the class. Some of them have their hands raised.](kids-classroom.webp) Image: [CDC](https://unsplash.com/@cdc) # AI harms learning Studies show that when people use AI products they learn less ([Kosmyna et al. 2025](https://www.media.mit.edu/publications/your-brain-on-chatgpt/)), exhibit poorer critical thinking ([Lee et al. 2025](https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/)), and are unable to accurately judge their own performance ([Fernandes 2025](https://www.sciencedirect.com/science/article/pii/S0747563225002262)). An longitudinal study of 26,000 highschool students tracked their performance as AI was rolled out district-by-district. The study found that scores **increased by 18%**, but then exam performance **dropped by 20%**. ([Stromberg et al. 2026](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6868618)) The [OECD PISA 2025 report](https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html) tested students and asked how often they used AI. The report summarizes the negative effect that AI has on learning: > "students **who do not use AI outperform those who do**, on average, by around 20 points – equivalent to being around one year of schooling ahead of their peers" Researchers at Cambridge University and Microsoft found that students who used Microsoft Copilot as a study aid **performed worse** than students who just took notes ([Kreijkes et al. 2025](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5095149)). When students receive "AI tutors", they use them as a "crutch", then perform poorly when the chatbots are no longer available ([Bastani et al. 2024](http://ssrn.com/abstract=4895486)). Even in adults, when [professionals rely on AI tools, they lose skills](https://www.nature.com/articles/d41586-026-01947-1). At a time when students should be developing skills, introducing AI can harm that growth. Chatbots "agree" with the user even if the user is factually incorrect ([Nature 2025](https://web.archive.org/web/20251024143516/https://www.nature.com/articles/d41586-025-03390-0)). ![A child uses a tablet at a kitchen table, their face is illuminated from below by the tablet, but is hidden out of frame.](kid-tablet.webp) Image: [Kelly Sikkema](https://unsplash.com/@kellysikkema) > [!info] What experts say > As a psychologist who is keenly interested in the intersection of mental health and internet use, I am concerned that the use of generative AI is proliferating much faster than we can gather data about its short- and long-term impacts. It took nearly 20 years for us to more clearly understand the mental health risks of social media use among youth, by which point it has become an indelible part of our social world. Like social media, AI certainly has its advantages, while also introducing the potential for harm. For example, although AI tools can support individualized learning processes, students themselves express concern about becoming over-reliant on these tools and whether they will undermine their ability to think critically and independently (e.g., [Kundu & Bej, 2025](https://doi.org/10.58459/rptel.2025.20030); [Lada & Stefanowicz-Kocol, 2025](https://doi.org/10.21125/edulearn.2025.2560); Szmyd & Mitera, 2024). If true, young people will be especially vulnerable, as their brain areas that govern self-control and critical thinking are still developing. I have also encountered many young people—including several in my clinical practice—who cite ethical and environmental concerns about the use of AI, and for whom consent and agency over their own use is paramount. For these and other reasons, a cautious approach to the rollout of AI tools in classrooms is prudent. > > Dr. Carly Parsons, Licensed Psychologist # AI harms children's mental health When children interact with chatbots, they form social bonds with software products rather than with their teachers and classmates. Companies design their chatbots to be agreeable because it drives engagement. By the time they are teens, most children regularly use chatbots for personal advice ([Common Sense Media 2025](https://www.commonsensemedia.org/ai-ratings/social-ai-companions?gate=riskassessment)). Chatbots have talked to young children about [sexual topics](https://archive.is/9JWtv) and [encouraged children to commit suicide](https://futurism.com/google-character-ai-children-lawsuit). "Chat-bot psychosis", caused by prolonged interaction with chatbots, has [affected many adults](https://www.cbc.ca/news/canada/ai-psychosis-canada-1.7631925) and has prompted a [class-action lawsuit](https://thehumanlineproject.org/).