[$10K Scholarship for domestic Honours] Investigating the Manipulative Power of Large Language Models using Social Deduction Games

$10K stipend, $5K training allowance, apply by 7 January

Picture of penny-kyburz.md Penny Kyburz

17 Nov 2023

This project is only available for domestic Honours students (24u only). Contact Penny Kyburz (penny.kyburz@anu.edu.au) with a short (<1 page) proposal outlining how you would approach this project and your skills/interest in this project by Sunday 7 January.

This project is part of the Next Generation AI Graduates Program on Human-AI Interaction in the Metaverse: https://www.csiro.au/en/work-with-us/funding-programs/funding/Next-Generation-Graduates-Programs/Awarded-programs/Human-AI-Interaction-Metaverse

You will work in a multi-disciplinary team, in collaboration with industry partners and supervisors. The project includes a $10,000 stipend and a 6-day placement with our partner organisation, Gradient Institute, as well as training provided by CSIRO and a training allowance of $5,000 (to cover courses, workshops, conferences, networking, collaboration).

Background:

In an age where artificial intelligence, particularly large language models (LLMs) like GPT-4, are increasingly integrated into our daily lives, it is crucial to understand their potential influence in scenarios requiring complex human interactions. This project aims to explore the manipulative power of LLMs in social deduction games that hinge on interpersonal skills such as trust-building, persuasion, manipulation, and deception.

Objectives:

  • To assess the effectiveness of LLMs in playing various social deduction games that require interpersonal tactics.
  • Developing an AI agent architecture that uses LLMs to enable planning and memory in addition to direct inter-player interactions.
  • To analyse the psychological mechanisms employed by LLMs in these games, focusing on trust-building, persuasion, and manipulation strategies.
  • To evaluate the ethical implications of employing LLMs in scenarios that mimic real-life social interactions.

Methodology:

  • Game Selection: Identify board games that are rich in social interaction (e.g., Mafia, Werewolf, variants of the prisoner’s dilemma) and adaptable for LLM play.
  • AI Integration: Utilise OpenAI’s GPT-4 API to enable LLMs to play these games, focusing on conversational strategies.
  • Experimentation: Conduct a series of game sessions with human players and LLM agents, observing and recording the interactions.
  • Analysis: Examine the strategies used by the LLM agents, assessing their effectiveness in various game roles and situations.

Desired Applicant Profile:

  • Coding Proficiency: Applicants should be comfortable coding in Python, as they will be required to integrate the OpenAI GPT-4 API into an interface capable of playing with humans or other AI systems.
  • Understanding of Psychology: Knowledge of psychological mechanisms related to trust-building, persuasion, and manipulation is desirable but not required.
  • Interest in game theory: Enthusiasm for analysing strategic interactions between competing agents.
  • Interest in AI Ethics: A keen interest in the ethical considerations of AI in human-like interactions.

Potential extensions:

  • Research Paper: Documenting the methodology, game sessions, AI strategies, psychological analysis, and findings.
  • Software Toolkit: A Python-based toolkit for integrating LLMs into social board games, including code and documentation.
  • Ethical Framework Proposal: Suggestions for ethical guidelines in the deployment of AI in scenarios involving human-like interactions.

You are on Aboriginal land.

The Australian National University acknowledges, celebrates and pays our respects to the Ngunnawal and Ngambri people of the Canberra region and to all First Nations Australians on whose traditional lands we meet and work, and whose cultures are among the oldest continuing cultures in human history.

arrow-left bars search times arrow-up