CS 222: AI Agents and Simulations

STANFORD UNIVERSITY, FALL 2024
Location: M W 01:30p-02:50p; Lathrop Library, Rm 299
Contact: cs222-ai-simulations@cs.stanford.edu

Final Project

Objective:

The goal of this project is to design and run a simulation using generative agents. We've compiled an agent bank that represents the members of the class, which you can choose to use if you wish. You will explore a research question using generative agents, drawing insights into behaviors, preferences, or interactions in a certain setting.

Project Overview:

Each group will use generative agents, either those that were created during the class or those you've created, to simulate behaviors and interactions. You’ll develop a research question around the capabilities and limitations of generative agents in simulating human behavior, such as decision-making, collaboration, or opinion diversity, and use the agents to conduct experiments, surveys, or social simulations that help answer that question.

Steps:

Team Formation:

Research Question Development:

Choose a research question that explores something meaningful about human behavior, agents, and/or simulations. A few examples that utilize the class agent bank might include:

Your research question should guide how you’ll use the generative agents to simulate interactions or responses, and should be situated in the context of prior work in that area. You should also consider how you will analyze the results of your simulation.

Simulation Design:

Design a simulation, survey, or experiment using the agents. Possible methods include:

Be creative! The project is open-ended. Feel free to ask us for help or to explore other simulations or experiments that you think might be interesting.

Running the Simulation:

If you choose to use the agents that represent each of your classmates, you’ll have access to the full agent bank, including interview data and agent models, which can be queried to simulate decisions or responses. Otherwise, you can create your own agents using the framework we've provided.

Record the outputs of the simulation (e.g., survey results, interaction logs) and analyze the findings in relation to your research question.

Analysis and Reflection:

Analyze the simulation outcomes. What did you learn through the behavior of the agents? How did the simulation reflect or differ from real-life dynamics?

Reflect on the effectiveness of the agents in representing their real-life counterparts and discuss any limitations or improvements.

Key Deliverables:

Project Proposal (10%):

Final Presentation (30%):

Final Submission (60%):

Grading Breakdown:

Important Dates:

Additional Considerations:

Optional Project Modifications:

If you have an alternative project idea that still utilizes the generative agents in a meaningful way, please check with your CA to ensure it meets the project goals.

Ethical Considerations:

Even though the generative agents are based on your classmates, ensure that your simulation and analysis maintain respect and privacy. Avoid sensitive topics or potentially uncomfortable scenarios.