A Nature Computational Science article discusses the use of large language models as human proxies, citing research on virtual humans, artificial societies, and language models as few-shot learners. Referenced works examine foundation models’ opportunities and risks, caution against using LLMs as human surrogates, and explore LLMs as simulated economic agents and tools for simulating human samples. Further cited studies cover generative agents, LLM-empowered agent-based modeling, whether AI language models can replace human participants, and surveys on LLM-based autonomous agents. The article also draws on classic work on similarity, emotion in believable agents, believable agents, artificial intelligence, multiagent systems, machine behaviour, and role play with large language models, as well as Argyle et al. on using LLMs for inference in the political and social sciences.