Don't Get Too Excited -- Eliciting Emotions in LLMs

2025-03-04

AI Summary

This study investigates the controllability of affective expression—specifically arousal and valence—in long-horizon, multi-turn dialogues generated by large language models (LLMs). Addressing the lack of systematic quantitative analysis in prior work, we propose a novel LLM-driven affective analysis framework integrated with multi-agent dialogue simulation: (1) generating multi-turn dialogue trajectories using open-source LLMs; (2) leveraging LLMs for self-supervised affective annotation; and (3) modeling and statistically analyzing affective dynamics within the arousal–valence space. Our experiments, the first of their kind, systematically reveal significant inter-model disparities—and shared bottlenecks—across mainstream LLMs in affective stability, capacity to generate extreme emotions, and contextual affective consistency (e.g., affective drift). The findings establish a reproducible evaluation benchmark and provide empirical grounding for designing affectively controllable LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

This paper investigates the challenges of affect control in large language models (LLMs), focusing on their ability to express appropriate emotional states during extended dialogues. We evaluated state-of-the-art open-weight LLMs to assess their affective expressive range in terms of arousal and valence. Our study employs a novel methodology combining LLM-based sentiment analysis with multiturn dialogue simulations between LLMs. We quantify the models' capacity to express a wide spectrum of emotions and how they fluctuate during interactions. Our findings reveal significant variations among LLMs in their ability to maintain consistent affect, with some models demonstrating more stable emotional trajectories than others. Furthermore, we identify key challenges in affect control, including difficulties in producing and maintaining extreme emotional states and limitations in adapting affect to changing conversational contexts. These findings have important implications for the development of more emotionally intelligent AI systems and highlight the need for improved affect modelling in LLMs.

Problem

• Assessing emotional expression in large language models.
• Evaluating affect control during extended dialogues.
• Identifying challenges in maintaining consistent emotional states.

Innovation

• Combines sentiment analysis with dialogue simulations
• Quantifies emotional expression range in LLMs
• Identifies challenges in maintaining emotional consistency