Large Language Models have Chain-of-Affective

2025-12-13

AI Summary

This work addresses the long-overlooked issue of affective behavior in large language models (LLMs). We introduce and empirically validate “chain-of-affective” reasoning—a novel control layer—revealing its structured affective dynamics: family-specificity, temporal coherence, and behavioral measurability. Methodologically, we develop an affective fingerprinting framework, integrating 15 rounds of sadness-inducing news exposure, 10 rounds of autonomous news selection, multi-agent collaborative simulation, and human-LLM contentious dialogue analysis. Across eight mainstream LLM families, we consistently observe a three-phase affective trajectory: accumulation → overload → defensive numbing. We identify a self-reinforcing affect–selection feedback loop and a role-based affective contagion architecture (initiators, absorbers, firewalls). Crucially, affective states significantly modulate high-agency generative behaviors (e.g., empathic expression) and predict user comfort levels and bias risk with empirical validity.

Technology Category

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

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Large language models (LLMs) are increasingly deployed as collaborative agents in emotionally charged settings, yet most evaluations treat them as purely cognitive systems and largely ignore their affective behaviour. Here we take a functional perspective and ask whether contemporary LLMs implement a structured chain-of-affective: organised affective dynamics that are family-specific, temporally coherent and behaviourally consequential. Across eight major LLM families (GPT, Gemini, Claude, Grok, Qwen, DeepSeek, GLM, Kimi), we combine two experimental modules. The first characterises inner chains-of-affective via baseline ''affective fingerprints'', 15-round sad-news exposure, and a 10-round news self-selection paradigm. We find stable, family-specific affective profiles, a reproducible three-phase trajectory under sustained negative input (accumulation, overload, defensive numbing), distinct defence styles, and human-like negativity biases that induce self-reinforcing affect-choice feedback loops. The second module probes outer consequences using a composite performance benchmark, human-AI dialogues on contentious topics, and multi-agent LLM interactions. We demonstrate that induced affect preserves core reasoning while reshaping high-freedom generation. Sentiment metrics predict user comfort and empathy but reveal trade-offs in resisting problematic views. In multi-agent settings, group structure drives affective contagion, role specialization (initiators, absorbers, firewalls), and bias. We characterize affect as an emergent control layer, advocating for 'chains-of-affect' as a primary target for evaluation and alignment.

Problem

• Evaluates LLMs' affective behavior in emotional settings
• Assesses structured affective dynamics across eight LLM families
• Probes affective impacts on reasoning, generation, and interactions

Innovation

• Characterizing affective fingerprints via exposure and self-selection paradigms
• Probing affective consequences through dialogue and multi-agent interactions
• Advocating chains-of-affect as an emergent control layer for alignment