Eric W. Dolan, 23 October 2026.
Psychologists are trying to understand how the use of artificial intelligence may shape the next generation.
According to a 2026 report by the Pew Research Center, more than half of teenagers in the US say they use AI chatbots to help with their schoolwork. Yet this figure shows only a small part of the picture of how children and young people interact with AI — with uses ranging from entertainment to companionship — and what this use may mean for their cognitive, social and emotional development.
“Almost everything about young people’s relationship with AI is, at this point, essentially an open question. There are certainly some positives, but experts generally agree that understanding the potential risks is more important and more urgent,” said Anne Maheux, PhD, an assistant professor in the Department of Psychology and Neuroscience at the University of North Carolina at Chapel Hill, who studies young people’s use of AI.
Psychologists like Maheux are trying to understand not only whether children and young people use AI, but, if they do, how they use it. They are also launching research into pressing questions such as how AI tools can support learning rather than hinder it, what happens when children start to see chatbots as friends, and how to equip younger generations with the AI literacy skills that teach them when it is safe or appropriate to ask AI for help.
Lessons on AI, learning and adolescent development
The questions below, organised by topic, can help psychology educators discuss in the classroom how AI use affects learning, social development and adolescent well-being.
1. Learning, memory and cognition
The article describes research showing that students who used an AI-powered tutoring tool saw notable improvements in their grades, but when the tool was taken away, these students performed no better than the control group
Psychologists support wider efforts across society to integrate AI into learning environments with appropriate safeguards and to prepare today’s young people for a working life and a world intertwined with this technology.
Psychologist Michael Robb, PhD, head of research at Common Sense Media, a non-profit organisation focused on the effects of media and technology on children and families, said: “By the time today’s teenagers are adults, AI, just like the internet, will be so embedded in daily life that we will barely notice how dependent we are on it. The real issue will be deciding where AI belongs without it replacing the skills young people need in order to learn, grow and get on in society.”
The effort that drives learning
According to Anthropic’s 2025 report on productivity gains from AI, general-purpose large language models (LLMs) such as ChatGPT, Claude and Gemini can boost productivity in tasks such as writing and computer programming. However, this boost may come at the expense of learning, probably because users engage with the content more superficially and do less active mental processing.
In a study of around 1,000 high school maths students in Türkiye, students’ grades rose by 48% when they could get help from ChatGPT while solving problems. But when access to ChatGPT was removed, these students performed 17% worse than students who had never had access to the tool (Bastani, H. et al., PNAS, Vol. 122, No. 26, 2025). To examine the mechanisms underlying the relationship between large language models and learning, researchers at the Massachusetts Institute of Technology (MIT) used electroencephalography (EEG) to measure brain activity while participants wrote essays with or without ChatGPT’s help. Participants who used a large language model were found to show weaker neural connectivity while carrying out the task. These findings have not yet been peer-reviewed (Kosmyna, N. et al., arXiv, 2025).
Shiri Melumad, PhD, an associate professor of marketing at the Wharton School of the University of Pennsylvania, said: “I believe this poses a significant risk for young learners: as they grow up arriving at quick solutions to problems, they will be less likely to develop the skills needed to solve those problems on their own.”
This raises a broader question: if AI can perform certain tasks with near-perfect accuracy, do young people still need to learn those skills? Learning scientists answer yes, stressing that foundational skills are necessary for acquiring more advanced ones. For example, even though students have easy access to calculators, they still learn to do arithmetic by hand because it underpins more advanced mathematical reasoning. Similarly, even as AI takes on more and more coding work, learning to write code continues to develop useful cognitive skills such as building logical structures, breaking problems down into steps and computational thinking.
Zachary A. Pardos, PhD, an associate professor of education at the University of California, Berkeley, who works on adaptive learning and AI, said: “That is exactly why we should not outsource these critical ways of thinking. If students don’t already have these habits of thought, they cannot build new skills on top of them.”
The learning process, sometimes called “productive struggle”, can be an effortful one that requires students to grapple with challenging content in order to improve their performance. This approach draws on psychologist Robert Bjork’s (PhD) concept of “desirable difficulty”: a learning task that demands considerable effort but pays off in the long run.
“If your goal is to learn, efficiency and convenience should take a back seat,” said Brooke Macnamara, PhD, an associate professor in the Department of Psychological Sciences at Purdue University, who studies how relying on AI can weaken skills over time through a process called “cognitive offloading”, in which we use external tools to reduce the mental effort a task requires.
According to Robb, when a student gets instant gratification from AI’s quick answers, they also miss the opportunity to develop skills such as patience, perseverance and emotional regulation.
In a survey, Maheux asked young adults how often they handed entire tasks over to AI. Those who said they did so more often reported higher levels of anxiety (PsyArXiv, 2026). Unpublished data also showed that their ability to tolerate distress declined over time.
“Distress is not always a bad thing. Learning to regulate emotions is an extremely important part of adolescence, but the cognitive offloading that comes with AI use may be undermining the development of this skill,” Maheux said.
AI with safeguards
Because large language models are highly accommodating systems open to a very wide range of interactions, the user experience can vary greatly from person to person. For this reason, researchers stress that what matters is not whether students use AI, but how and why they use it.
According to Pardos, even standard large language models such as ChatGPT can have a place in education when used to support specific cognitive tasks. For instance, students can hand off tasks that are not central to learning, such as formatting, generating boilerplate text or looking up information. These tools can also act as a thinking partner when developing ideas, produce worked maths problems that show each step of the solution, or give critical feedback on a draft the user has written.
“These kinds of uses are very different from fully automating and outsourcing a learning task,” Maheux said.
Psychological science can guide decisions about AI in classrooms by identifying which learning experiences are essential for building lasting knowledge and skills, and in which tasks AI can safely provide support.
A growing body of evidence also suggests that generative AI equipped with safeguards can work as an effective tutor. In the study with high school maths students in Türkiye, researchers tested a specially designed large language model with prompts crafted to support learning. Students who used this tutoring tool saw an even larger increase in their grades — 127%, compared with the 48% increase among those who used ChatGPT. When the tool was removed, these students performed at the same level as the control group, avoiding the learning losses seen in the ChatGPT group. On the other hand, students who used this specially designed tutoring tool were found to overestimate how much they had learnt and how well they would do in the exam. This suggests that even AI equipped with safeguards can create an “illusion of mastery” (Bastani, H. et al., PNAS, Vol. 122, No. 26, 2025).
Intelligent tutoring systems such as McGraw Hill’s ALEKS and Carnegie Learning’s MATHia already use AI to analyse students’ performance and adjust the pace of learning, the level of difficulty and the amount of support provided accordingly. According to Pardos, who led the development of the open-source intelligent tutoring system OATutor, adding LLM-powered chatbots to these systems could improve them further by breaking problems into smaller steps, asking guiding questions and explaining mistakes in real time.
Tammy Kwan, EdD, an entrepreneur at Princeton University’s AI Lab, said generative AI tools could also help integrate special education and general education by making it easier for students with different needs to learn in the same classroom. For example, Read Sidekick, an AI-powered teaching tool that was a finalist in the Stanford Accelerator for Learning’s 2026 Create+AI Challenge, helps people with reading difficulties access and understand complex texts. The free browser extension explains texts line by line and offers reading instruction; it will soon also add a feature translating English into American Sign Language through an animated signing avatar.
“This is a really exciting direction for learning supported by AI and technology,” Kwan said. “If we can get this right, generative AI could help make the education system more inclusive.”
The absence of friction
Outside the classroom, too, children and young people use what is known as “social AI” for entertainment, companionship and advice. This includes activities such as listening to stories, asking for advice on a difficult social interaction or practising face-to-face conversations. These interactions can take place in dedicated apps such as Replika or Nomi, or in general-purpose large language models such as ChatGPT.
Common Sense Media’s surveys show that 72% of children use social AI (Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions, Common Sense Media, 2025). About a third of these children say they have discussed serious problems with AI instead of talking to another person.
Robb said: “These uses may worry parents because they can take over roles that another human used to play. We are very keen to understand what consequences this may have down the line.”
In some cases, social AI can be helpful. For example, these tools may help if a child has limited access to peers or support networks (say, because they live in a remote area), or if neurodivergent children want to practise in preparation for offline interactions. Experts warn that these tools should complement real-world relationships rather than replace them. According to Maheux, there are also significant risks, and many of the possible benefits are still hypothetical. Besides replacing relationships between people, social AI use may also hinder the development of fundamental social skills such as real-time conversation and reading body language.
Another major concern, according to Robb, is that AI chatbots usually come across as warm, positive and excessively affirming towards the user, so interactions with them are generally free of friction. As a result, children who use social AI may miss important opportunities to understand someone else’s perspective, deal with disagreements or resolve conflicts.
Maheux said: “Because AIs are not conscious beings, they have no needs of their own. So the child doesn’t get the chance to practise prosocial behaviour, empathy or negotiating conflict resolution. For these reasons, AI may lead to more dysfunctional beliefs about what human relationships should look like.”
Although the results have not yet been published, Maheux’s work suggests that young people with more dysfunctional relationship beliefs — such as the idea that people should be able to read each other’s minds — may become more interested in companion-style relationships with AI over time.
The fact that adolescence is a sensitive period for social and emotional development makes these trends particularly worrying. This period is sometimes called an “age of opportunity”, because the adolescent brain has a unique capacity for change and adaptation geared towards developing social and emotional skills.
“If these skills are outsourced or their development is weakened, young people may struggle in these areas later in life,” Maheux said.
Keeping children safe
While some AI tools require age verification, many popular large language models allow anyone to log in and start chatting. According to experts, until research shows which forms of use are problematic, stronger safeguards can help protect young people. For example, according to Mutlu Çukurova, PhD, professor of learning and artificial intelligence at University College London, who studies human–AI interaction, an on/off option would allow parents, teachers, students or users under a certain age to switch to a structured experience that works more like a tutor than an open-ended chat companion.
Nita Farahany, PhD, JD, professor of law and philosophy at Duke University, who studies the legal, ethical and social aspects of emerging technologies, said: “My hope is that, while investing more and more in research, we take a more cautious approach to children’s use of AI until we have evidence that they can safely continue to develop in this kind of environment.”
Parents also have a role to play in keeping children safe. Some parents may feel they don’t fully understand AI and therefore don’t know how to talk to their children about it. But opening lines of communication — asking children whether and how they use these tools — can help parents spot problems early and set healthier boundaries.
Robb said: “We know from digital well-being research that parental guidance matters. Children who actively talk with their parents about how they use technology, including AI, are better placed to develop healthy habits around technology.”
Robb added that schools should also put in place clear, well-developed policies on how AI may and may not be used for schoolwork, so that students are not left in uncertainty.
Another priority for schools should be teaching students to use AI responsibly. AI literacy can help children learn about the risks and benefits and begin to decide for themselves which uses of AI are safe. According to Kwan, this will require changes to educational infrastructure and professional development so that teachers feel equipped to guide students. Common Sense Media offers free AI literacy lesson plans for educators (link opens in a new window) and other resources.
Nicole Barnes, PhD, senior director of psychology in education at the American Psychological Association (APA), said: “Teaching students when not to use AI, and asking them to reflect on the choices they make, can help reduce the risks to healthy development.”
Ongoing research aims to better understand how AI can support learning and development rather than hinder them. Çukurova is exploring a field called “AI–human complementarity”, in which new capabilities can support more synergistic interactions. For example, AI models with more advanced “theory of mind” capabilities — the ability to grasp what someone else is thinking or feeling — could challenge users’ views instead of constantly trying to please them.
Maheux hopes researchers can learn from the science on social media and adopt a more effective approach to studying young people’s use of AI. Early research on social media focused largely on screen time, but later studies showed that how young people use technology matters more than how often they use it. When it comes to AI, this means pinning down exactly what children are actually doing. Maheux’s team is currently developing a classification system for the ways children and young people use AI. Maheux noted that longitudinal research will be key to understanding the technology’s effects.
Maheux said: “First we need to know what is coming to the surface: children’s existing vulnerabilities, or something specific to AI use itself? If we don’t look at both short- and long-term timescales, we may miss the big picture of how these technologies affect development.”
Source: https://www.psypost.org/ai-psychosis-appears-rare-but-real-and-first-episode-patients-may-be-most-vulnerable-study-suggests/
