Social Simulacra: Creating Populated Prototypes for Social Computing Systems
The paper introduces "social simulacra," a prototyping approach that generates large, realistic synthetic populations of users, posts, and replies to explore the behavior of social computing systems at scale. The authors present SimReddit, a tool that leverages GPT-3 prompt chains to automatically populate Reddit‑style communities (GENERATE), simulate alternative conversation branches or moderator interventions (WHATIF), and resample divergent outcomes (MULTIVERSE). Technical evaluation recreates 50 real subreddits, showing that human judges often cannot reliably distinguish synthetic from real content. A designer study with 16 participants demonstrates that simulacra help identify edge cases, refine community rules, and anticipate moderation challenges. The paper discusses limitations, ethical concerns (e.g., generation of offensive content), and situates the work within a broad literature on online harassment, language model prompting, prototyping methods, and social media research.
Topics
Social Simulacra: Concept and Foundations
Introduces the overarching idea of creating fully populated synthetic communities—social simulacra—to prototype and evaluate social computing systems before deployment. Situates the approach within prior work on prototyping methods, social media analysis, and language‑model prompting.
SimReddit: GPT‑3 Prompt Chains for Synthetic Communities
Describes the SimReddit tool architecture, which uses chained GPT‑3 prompts to (1) GENERATE realistic Reddit‑style posts and replies, (2) simulate alternative conversation branches or moderator interventions (WHATIF), and (3) resample divergent outcomes (MULTIVERSE).
Evaluation: Reconstruction of Real Subreddits and Designer Study
Details the two‑pronged evaluation: (a) technical reconstruction of 50 real subreddits, showing human judges often cannot distinguish synthetic from real content, and (b) a qualitative study with 16 designers who used simulacra to refine community policies and anticipate moderation challenges.
Ethical and Societal Considerations: Harassment, Bias, and Misinformation
Places the work within literature on online harassment, fake news, and language‑model bias, discussing potential harms of synthetic content generation and proposing mitigation measures.