Artificially intelligent agents in the social and behavioral sciences: A history (arxiv.org)

This paper traces the evolution of “agentic” AI—software agents used to model, experiment on, or behave like social actors—within the social and behavioral sciences from the emergence of programmable computers to today’s large language models. It highlights early social simulations and the difficulty of communicating their value in a pre-computer public, the rise of social-systems and game‑theoretic agent models, the methodological shock of big data, and the recent pivot to generative AI where LLMs are employed both as experimental tools and as subjects in behavioral studies. The authors emphasize that AI has not just provided new tools but reshaped the scientific process, from hypothesis generation and simulation to data collection, interpretation and replication. For the AI/ML community the review is significant because it frames technical advances (agent architectures, game-theoretic agents, simulation platforms, and LLMs) in an epistemic and methodological context: agent-based models enable controlled “virtual experiments,” big data created new inference opportunities and biases, and generative models raise questions about validity when agents mimic human responses. The paper underscores practical implications—need for open code/data/demos, careful validation of agent behavior, interdisciplinary collaboration, and attention to reproducibility, ethics, and how entwined modeling tools are with the phenomena they study.