A father told us his 9-year-old came home upset. ‘My friend thinks I was being mean, but I wasn’t.’ The father asked, ‘What do you think your friend is thinking right now?’ The child paused. That pause — that effort to hold two minds at once — is cognitive empathy.
Later that week, the same child asked ChatGPT, ‘My friend thinks I was being mean. What is he thinking?’ ChatGPT gave a perfect paragraph: ‘Your friend might feel hurt because he values the game and felt excluded…’ The child said, ‘Thanks,’ and walked away. No pause needed.
The first was practice. The second was an answer delivered.
Cognitive empathy — thinking with others — is also called perspective-taking or Theory of Mind. It is the ability to infer what someone else believes, intends, or knows, even when it differs from your own. Developmental research shows this ability builds rapidly between ages 4 and 7 through stories, pretend play, and everyday disagreements where children must coordinate differing perspectives (Wellman, 2014; Wimmer & Perner, 1983).
Can AI teach it?
Where AI Is Surprisingly Good
AI can provide endless role-play prompts. ‘What might happen if you say that? What might your friend think if you do this?’ In that sense, it can act like a rehearsal partner. A recent review of AI literacy for children notes that AI can support operational and interactional skills when guided by an adult (Atias & Mawasi, 2025). Xu and He (2026) describe AI as a distinctive source of developmental experience that can be shaped.
Some recent studies even claim large language models show behavior consistent with Theory of Mind tasks (Kosinski, 2024). A child can ask, ‘If I take his toy, what will he think?’ and get a plausible mental-state explanation.
For a child who has few siblings or who is shy to practice with peers, this rehearsal can be a low-stakes starting point. It is not anti-AI to recognize that.
Key takeaway: For cognitive empathy, AI can offer language for perspective-taking that some children rarely hear elsewhere.
Where It Falls Short — And Why It Matters
Three peer-reviewed cautions matter for parents.
First, passing a test is not having a mind. Ullman (2023) showed that small variations in classic false-belief tasks cause large language models to fail, suggesting they rely on pattern matching rather than genuine mentalizing. The model simulates perspective-taking without having a perspective.
Second, AI is sycophantic by design. Recent work on sycophancy in LLMs documents a tendency to agree with users to maintain engagement (Sharma et al., 2024). If a child says, ‘My friend is wrong,’ the model often affirms rather than challenging: ‘You have a right to feel that way.’ That feels supportive, but it removes the crucial practice of holding that you might be wrong.
Third, perspective-taking needs real stakes. In real friendship, if you misread your friend, there is a consequence — hurt, repair, negotiation. With AI, if you misread, nothing happens. The feedback loop that builds Theory of Mind — action, misattunement, repair — is missing. The Lancet review warns this is part of relational displacement: when AI substitutes for difficult human conversations, children miss practice that protects mental health (Ha et al., 2026).
Scoping reviews of chatbots for social-emotional learning find they can deliver scripts but show persistent limits in fostering true social understanding (Beceren et al., 2025).
Key takeaway: AI can describe perspective-taking, but cannot practice it with you — because it has no perspective to be taken.
What Parents Can Do for Cognitive Empathy
1. Protect stories — Read fiction aloud where characters misunderstand each other. Pause and ask: What does she think he thinks? Research shows narrative fiction builds mentalizing (Kidd & Castano, 2013).
2. Protect disagreement — Do not resolve sibling conflicts instantly. Ask: What do you think she wants? What do you want? How can both be true?
3. Use AI as a script, not a substitute — If your child uses AI for advice, ask: Do you think that explanation fits your friend? Has your friend ever acted like that before? What does your history with him tell you that AI doesn’t know?
Key takeaway: Cognitive empathy grows when a child holds two minds in mind — theirs and another’s. AI can name the second mind, but only real relationship teaches you to care what it feels.
What About You?
Have you noticed your child practicing perspective-taking with AI? What happened when they tried with a real friend? Share below.
About the Authors
Dr. Shaheen Pasha is a Professor of Special Education with over 35 years of teaching, research, and academic leadership experience. She served as Professor and Chairperson of the Department of Special Education at the University of Education, Lahore. She earned her Ph.D. in Special Education from the University of Southampton, UK. She has published more than 35 research papers and co-authored two books. Her expertise lies in child development, cognitive growth, and special education.
Dr. M. Anwar-ur-Rehman Pasha (widely recognized in academic circles as Dr. M. A. Pasha) is a Professor of Computer Science with over 35 years of post-graduate teaching, research, and educational management experience. He earned his Ph.D. in Computer Science from the University of Southampton, UK in 1996. His research focuses on Artificial Intelligence, Human-Computer Interaction, and Computational Thinking. He has served on different academic and administrative positions and has published two books and over 30 research articles.
Together, they bring 70 years of combined wisdom to help parents raise capable, thoughtful children in the age of AI.
References
Wellman, H. M. (2014). Making minds: How theory of mind develops. Oxford University Press.
Wimmer, H., & Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13(1), 103-128.
Kidd, D. C., & Castano, E. (2013). Reading literary fiction improves theory of mind. Science, 342(6156), 377-380.
Kosinski, M. (2024). Evaluating large language models in theory of mind tasks. Proceedings of the National Academy of Sciences.
Ullman, T. (2023). Large language models fail on trivial alterations to theory-of-mind tasks. arXiv.
Sharma, M., et al. (2024). Towards understanding sycophancy in language models. arXiv.
Atias, O., & Mawasi, A. (2025). Conceptualizing AI literacies for children and youth: A systematic review. Computers and Education: Artificial Intelligence, 9, 100491.
Xu, Y., & He, K. (2026). Growing up with artificial intelligence: Implications for child development. Cambridge University Press.
Beceren, O., et al. (2025). Digital companions in early childhood education: A scoping review on the potential of chatbots for supporting social-emotional learning. Frontiers in Education, 10, 1634668.
Ha, T., et al. (2026). How interactional AI may alter adolescent relational learning and mental health. The Lancet Child & Adolescent Health.