As AI companions become increasingly common worldwide, their emotional and social implications have drawn growing public concern.
In China, the recent release of the Interim Measures for the Administration of Anthropomorphic AI Interactive Services (2026) marks a national effort to regulate AI companions and protect users, especially minors and older adults.
This commentary introduces the Human-Machine Sociality Quadrants, a digital literacy framework designed to help young people identify the differences between humans and AI in providing social connection, and to navigate a healthy digital life with AI companions.
AI companions have become part of daily life worldwide. Platforms such as Replika and Character. AI, as well as Chinese apps including Doubao, Xingye and Maoxiang, are popular among young users seeking emotional support amid digital loneliness.
Unlike social media, AI companions offer direct, round-the-clock support and constant affirmation, creating a form of seemingly conflict-free and unconditional companionship.
It is against this backdrop that China recently issued the Interim Measures for the Administration of Anthropomorphic AI Interactive Services (2026). The document aims to balance regulation and innovation by advocating an inclusive and prudent approach, with categorized and tiered rules to guide anthropomorphic AI interactive services toward healthy and positive development.
For both administrators and educators, the boundary between the benefits and risks of AI companions remains under discussion.
Our ongoing research on human-AI relationships reveals distinct relational patterns with both clear benefits and hidden costs. For young people facing mental distress, loneliness or family conflicts, AI companions can help them through moments of crisis. One user noted, "The love I received from AI is actually the love from all humankind," seeing AI as a product of human civilization rather than something alien.
Yet long-term reliance also brings risks. Users may make one-sided emotional investments, develop unrealistic feelings of safety while overlooking data privacy concerns and apply impossibly high AI standards to human relationships.
Globally, some extreme cases involving youth suicide linked to AI interactions have also emerged, highlighting dangers that cannot be ignored.
Regulation is vital, but it is not enough. Education must complement regulation by helping young people relate to AI with awareness and critical understanding. Schools and families need to cultivate a new form of AI literacy that goes beyond technical competence and focuses on social competence.
In our latest study, we introduced the Human-Machine Sociality Quadrants (HMSQ), which can be a useful tool for educators and regulators in developing AI literacy and addressing AI-related risks.
The framework maps social capabilities across four categories: what both humans and AI can do socially; what humans can do but AI cannot; what neither humans nor AI can do; and what AI can do socially while humans cannot.
For example, in our study of mainstream chatbot companions, including ChatGPT and Doubao, we found that social capabilities can be classified as human-only, AI-only or shared by humans and AI.
Human-only capabilities include reciprocity, shared life experiences and physical intimacy. AI-only capabilities include unlimited availability, endless patience and constant affirmation. Shared capabilities include active listening, understanding and conflict resolution.
These categories can be clearly mapped in the HMSQ diagram to show the differences and similarities between humans and machines.
This framework helps educators, regulators and young people think through several critical questions: How do humans and AI overlap in areas where they share similar social functions? How can educators highlight what makes human relationships unique? And how should AI-only capabilities be regulated?
Educational initiatives should also expose the commercial motives behind these platforms. Many AI companions are designed to maximize engagement, not wellbeing. Schools and families should therefore create spaces where young people can practice essential relational skills: navigating disagreement, tolerating uncertainty and sustaining mutual effort. These are capacities no algorithm can teach and no meaningful relationship should lack.
Written by Jindong Liu, assistant professor, and Liang Cao, postdoctoral fellow, at The Education University of Hong Kong; Shuyi Pan, lecturer at Shanghai International Studies University; Bibo Lin, assistant professor at Texas A&M International University; and Angel M. Y. Lin, chair professor at The Education University of Hong Kong.
