Simulating human behavior with AI agents promises a testbed for policy and the social sciences. We interviewed 1,000 people for two hours each to create generative agents of them. These agents replicate their source individuals’ attitudes and behaviors. 🧵

RT @oshaikh13: What if LLMs could learn your habits and preferences well enough (across any context!) to anticipate your needs?. In a new p….

What do you want to know?.

RT @StanfordHAI: 📢 New policy brief: AI agents that can simulate human behaviors and attitudes can help test ideas in social science. Our l….

RT @dorazhao9: Todo lists, docs, email style – if you've got individual or team knowledge you want ChatGPT/Claude to have access to, Knoll….

RT @msbernst: Book announcement: with @stanfordmav, we are publishing "Flash Teams: Leading the Future of AI-Enhanced, On-Demand Work." It'….

RT @tizianopiccardi: New paper: Do social media algorithms shape affective polarization?. We ran a field experiment on X/Twitter (N=1,256)….

RT @emollick: Crazy interesting paper in many ways:.1) Voice-enabled GPT-4o conducted 2 hour.interviews of 1,052 people.2) GPT-4o agents we….

RT @merrierm: Excited to share this new work that shares a new method for creating realistic generative agents that can be used for synthet….

RT @cqzou: So thrilled to share our work on this~!. Grounding generative agents in real, verifiable behavior is the step that takes this me….

RT @msbernst: In a follow-up to the generative agents paper, @joon_s_pk+team demonstrate that anchoring agents in rich qualitative informat….

RT @percyliang: How close can LM agents simulate people? We interview person P for 2 hours and prompt an LM with the transcript, yielding a….

RT @manoelribeiro: This is groundbreaking, super exciting work!.

Thank you to my coauthors, @msbernst, @percyliang, @RobbWiller, @cqzou, @aaronshaw, @makoshark, @merrierm, @carriejcai. And thank you @KolluriAkaash for helping out with the open source release, and to @StanfordHCI and @StanfordNLP for fostering this work. (14/14).

In closing, doing great interdisciplinary work that respects the tradition and rigor of each field is beyond any one person. This work would not have been possible without an all-star team that embodied its interdisciplinary nature, intersecting AI and social sciences. (13/14).

For those interested, here is an open-source repository and a Python package for this work:.Github: (While we are not releasing the participant data, I have included my personal generative agent in the repo. :)) (12/14).

So, to support research while protecting participant privacy, we (Stanford authors) plan to offer a two-pronged access system in the coming months: 1) open access to aggregated responses on fixed tasks, and 2) restricted access to individual responses on open tasks. (11/14).

We spent countless hours discussing ethics with the team, the IRB, and participants. Here’s what we believe: systems hosting generative agents of real people must, at a minimum, support usage audits, provide withdrawal options, and respect individuals' consent and agency. (10/14).

At the same time, this work points to the beginning of an era in which generative agents can represent real people. This ought to bring both excitement and concerns: how can we balance the potential benefits while safeguarding individuals' representation and agency? (9/14).

In sum, this work opens the door to simulating individuals. We believe that accurately modeling the individuals who make up our society ought to be the foundation of simulations. The resulting agent bank of 1,000 generative agents will further facilitate this function. (8/14).

In addition, our interview-based agents reduce accuracy biases across racial and ideological groups compared to agents provided with demographic descriptions. We attribute this to the agents in our study reflecting the myriad idiosyncratic factors of real individuals. (7/14).

Our finding: the agents perform well. They replicate participants' responses on the General Social Survey 85% as accurately as participants replicate their own answers two weeks later, and perform comparably in predicting personality traits and experimental outcomes. (6/14).