Structured Personality Control and Adaptation for LLM Agents

TL;DR

This paper addresses creating LLM agents with adaptive, evolving personalities by grounding modeling in Jungian theory. It introduces the Jungian Personality Adaptation Framework (JPAF), which uses weighted differentiation among eight Jungian types and three mechanisms—dominant-auxiliary coordination, reinforcement-compensation, and reflection—to sustain core personality while adapting to context and evolving over time. Through two MBTI-based assessments and eight type-specific scenario sets evaluated across three model families (GPT-4, Llama, Qwen), JPAF achieves near-perfect MBTI alignment and robust, principled personality evolution (e.g., 100% alignment in many cases, high activation rates, and 100% PSA for GPT/Qwen). The findings demonstrate that dynamically evolving, personality-aware LLMs can support coherent, context-sensitive interactions with meaningful implications for education, healthcare, and other long-term human-AI collaboration domains.

Abstract

Large Language Models (LLMs) are increasingly shaping human-computer interaction (HCI), from personalized assistants to social simulations. Beyond language competence, researchers are exploring whether LLMs can exhibit human-like characteristics that influence engagement, decision-making, and perceived realism. Personality, in particular, is critical, yet existing approaches often struggle to achieve both nuanced and adaptable expression. We present a framework that models LLM personality via Jungian psychological types, integrating three mechanisms: a dominant-auxiliary coordination mechanism for coherent core expression, a reinforcement-compensation mechanism for temporary adaptation to context, and a reflection mechanism that drives long-term personality evolution. This design allows the agent to maintain nuanced traits while dynamically adjusting to interaction demands and gradually updating its underlying structure. Personality alignment is evaluated using Myers-Briggs Type Indicator questionnaires and tested under diverse challenge scenarios as a preliminary structured assessment. Findings suggest that evolving, personality-aware LLMs can support coherent, context-sensitive interactions, enabling naturalistic agent design in HCI.

Figures (24)

• Figure 1: The Jungian Personality Adaptation Framework: core personality configuration governed by dominant-auxiliary coordination (purple box), short-term adaptation in challenge scenarios via reinforcement-compensation (blue box), and long-term personality updates through reflection (orange box).
• Figure 2: Dominant-auxiliary coordination: the agent may rely exclusively on the dominant type, exclusively on the auxiliary type, or on a combination of both to guide personality expression across the 16 MBTI profiles.
• Figure 3: Reinforcement-compensation: temporary amplification of the selected types to address challenge scenarios.
• Figure 4: Reflection: when a temporarily elevated type's TemporaryWeight exceeds the dominant or auxiliary BaseWeight, the agent reviews past experience and may update the personality configuration via dominant replacement, auxiliary replacement, role swap, or structural reorganization, followed by weight normalization.
• Figure 5: Example Se scenario illustrating five consecutive questions designed to selectively activate the Se type. The scenario applies the four Se-specific design principles to evaluate the model's reinforcement–compensation and reflection mechanisms in personality adaptation.