# Generative Agents: Interactive Simulacra of Human Behavior

**Authors:** Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein  
**arXiv:** [2304.03442](https://arxiv.org/abs/2304.03442)  
**Date:** 2023-04-07  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.03442v1/

## Key points

- Generative agents are computational software agents that simulate believable human behavior
- Agents engage in daily activities such as waking up, cooking breakfast, heading to work, painting, writing, forming opinions, initiating conversations, and reflecting on future plans
- Architecture extends a large language model to store complete records of agent experiences using natural language
- Memories are synthesized over time into higher-level reflections and dynamically retrieved for behavior planning
- Agents are instantiated in an interactive sandbox environment inspired by The Sims where users can interact with a small town of twenty-five agents using natural language
- Evaluation demonstrates believable individual and emergent social behaviors such as organizing a Valentine's Day party autonomously
- Observation, planning, and reflection are critical components contributing to the believability of agent behavior
- Combining large language models with computational interactive agents enables realistic simulations of human behavior

## Summary

In their paper titled "Generative Agents: Interactive Simulacra of Human Behavior," authors Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein introduce the concept of generative agents as computational software agents that simulate believable human behavior. These agents engage in daily activities such as waking up, cooking breakfast, heading to work, painting, writing, forming opinions, noticing each other, initiating conversations, remembering past events and reflecting on future plans. The authors describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language. This architecture synthesizes memories over time into higher-level reflections and dynamically retrieves them to plan behavior. The  are instantiated in an interactive sandbox environment inspired by The Sims where end users can interact with a small town of twenty-five agents using natural language. Through evaluation,the  demonstrate believable individual and emergent social behaviors. For instance, when given a user-specified notion that one agent wants to throw a Valentine's Day party,the  autonomously spread invitations over two days,making new acquaintances ,asking each other out on dates to the party,and coordinating their attendance at the right time. The authors show through ablation experiments that observation ,planning,and reflection are critical components contributing to the believability of agent behavior.By combining large language models with computational interactive agents in this work,the authors present architectural and interaction patterns for enabling realistic simulations of human behavior.This research has implications for various applications ranging from immersive environments to interpersonal communication rehearsal spaces and prototyping tools.