Risks of counter-stereotypical AI when it replaces human contact

Artificial intelligence (AI) agents are often designed with stereotype-consistent social cues-such as feminized voice assistants in service roles or masculine security robots-to increase acceptance and interaction fluency. Yet these same design choices may reinforce the stereotypes they draw on. Recent work, however, proposes a more optimistic possibility: counter-stereotypical AI-for example, a female-coded technical chatbot or male-coded care robot-may reduce prejudice by functioning as synthetic intergroup contact, repeatedly exposing users to agents that serve as proxy members of social groups and challenge familiar stereotypes. We build on this account but argue that its implications depend critically on whether AI supplements or replaces human interaction. When counter-stereotypical AI adds exposure to underrepresented groups, it may challenge stereotypes by broadening the range of people users associate with particular traits or domains. When it substitutes for humans, however, its effects become more ambivalent. We identify three risks of counter-stereotypical AI as replacement: non-integration, when users treat the agent as a designed persona rather than revising beliefs about people; substitution, when synthetic contact displaces the reciprocal and agentic qualities of human contact; and illusion of inclusion, when visible AI representation is mistaken for institutional inclusion. We position replacement by counter-stereotypical AI as a critical boundary condition for theories of synthetic contact and representational AI, and as a key context for future research on prejudice, technology, and social change.