PsySpace: Simulating Emergent Psychological Dynamics in Long-Duration Space Missions using Multi-Agent LLMs

Ibrahim Khan, Mustafa Can Gursesli, Ruck Thawonmas

Keywords: LLMs, Psychology, Space, Simulations

TL;DR: This paper presents PsySpace, a novel multi-agent framework that uses Large Language Models to simulate the psychological dynamics of astronaut crews.

Abstract: This paper presents PsySpace, a novel multi-agent framework that uses Large Language Models (LLMs) to simulate the emergent psychological dynamics of astronaut crews on long-duration space missions. Current methods for studying space psychology, such as analog missions, are resource-intensive and not scalable. To address this, we introduce agents with a dual-component psychological architecture, comprising a static personality profile and a dynamic state vector for stress and loneliness that evolves based on interactions within a data-driven mission environment. We demonstrate that PsySpace can replicate complex psychological phenomena observed in real-world missions, such as the ``third-quarter'' effect. Furthermore, we introduce an AI-based Psychological Support Agent (PSA) and show through bootstrapped A/B testing that its interventions cause a statistically significant reduction in crew stress. Our comparative analysis of five different LLM architectures reveals distinct behavioral fingerprints, establishing a new benchmark for evaluating the social intelligence of generative agents. We believe PsySpace provides a powerful, low-cost tool for enhancing mission planning, crew selection, and the development of AI to support human well-being in high-stakes environments.

Submission Number: 293