Generative Agents: Interviews, Surveys, Behavioral Tasks, and Digital Representations of 1,052 U.S. Adults
Creators:
Joon Sung Park; Carolyn Q. Zou; Jonne Kamphorst; Niles Egan; Aaron Shaw; Benjamin Mako Hill; Carrie Cai; Meredith Ringel Morris; Percy Liang; Robb Willer; Michael S. Bernstein
Publication Date:
2024-11-15
Data Category:
Dataset Description:
| Important features / special characteristics: Digital representations of a diverse national sample of 1,052 U.S. adults, grounded in sensitive real-person self-reports. Each participant completed an approximately two-hour AI-conducted semi-structured interview (about 6,500 words per person), structured surveys, and behavioral tasks. Agents were evaluated using General Social Survey items, the 44-item Big Five inventory, five behavioral economic games, and five social-science experiments; participants repeated tasks two weeks later for a test–retest benchmark. Data size: Not reported in GB. Number of observations: 1,052 participants / agent representations. The exact number of row-level records is not reported. Temporal coverage: Study collection dates are not reported in the cited public materials; evaluation includes a two-week repeat-measurement interval. Structure: Three principal input/configuration groupings are described: (1) two-hour interview transcripts, (2) structured survey responses, and (3) combined interview-plus-survey inputs. Evaluation data cover GSS responses, Big Five personality items, economic games, and social-science experiments. Important classification note: The digital agents and their generated predictions are synthetic, but the grounding interviews, surveys, and task responses are real individual-level self-report data. |
Variables:
Details: