Persona Dynamics: Unveiling the Impact of Persona Traits on Agents in Text-Based Games

ACL ARR 2024 December Submission1803 Authors

Abstract: Artificial agents play an increasingly integral role in interactions and decision-making, yet aligning them with human value remains a significant challenge. This paper investigates how personality traits influence agent performance in text-based adventure games. We introduce PANDA: Personality-Adapted Neural Decision Agents, a methodology for projecting human personality traits onto agents to guide their behavior. To induce personality in a text-based game agent, we employ two key steps: (1) training a personality classifier to determine an agent’s personality type, and (2) integrating this classifier into the agent’s learning process. We deployed agents embodying 16 distinct personality types across 25 text-based games and analyzed their trajectories, we could demonstrate that (1) AI as an action-oriented agent can be steered toward specific personalities, and (2) certain personalities, notably openness, confer advantages in the interactive game environment.

Paper Type: Long

Research Area: Human-Centered NLP

Research Area Keywords: human factors in NLP

Contribution Types: Model analysis & interpretability

Languages Studied: English

Submission Number: 1803