On Emotion-Sensitive Decision Making of Small Language Model Agents
Summary: arXiv:2604.06562v1 Announce Type: new
Abstract: Small language models (SLM) are increasingly used as interactive decision-making agents, yet most decision-oriented evaluations ignore emotion as a causal factor influencing behavior. We study emotion-sensitive decision making by combining representation-level emotion induction with a structured game-theoretic evaluation. Emotional states are induced using activation steering derived from crowd-validated, real-world emotion-eliciting texts, enabling controlled and transferable interventions beyond prompt-based methods. We introduce a benchmark built around canonical decision templates that span cooperative and competitive incentives under both complete and incomplete information. These templates are instantiated using strategic scenarios from Diplomacy, StarCraft II, and diverse real-world personas. Experiments across multiple model families in various architecture and modalities, show that emotional perturbations systematically affect strategic choices, but the resulting behaviors are often unstable and not fully aligned with human expectations. Finally, we outline an approach to improve robustness to emotion-driven perturbations.
Introduction
The advent of small language models has revolutionized the way we interact with artificial intelligence. These models are not only capable of generating human-like text but are increasingly being utilized as decision-making agents. However, a significant gap exists in the understanding of how emotions influence their decision-making processes.
Research Focus
This study aims to bridge that gap by examining emotion-sensitive decision-making in small language models. Traditional evaluations have overlooked emotions as a critical factor in determining behavior, which this research seeks to address.
Methodology
Our approach combines two primary elements:
- Emotion Induction: We employ activation steering to induce emotional states in the models using texts validated by crowdsourcing that are designed to elicit real-world emotions.
- Game-Theoretic Evaluation: We utilize a structured evaluation framework that assesses decision-making through a series of canonical templates that represent both cooperative and competitive scenarios.
Benchmark Development
We introduce a benchmark that incorporates various decision-making templates based on established games such as:
- Diplomacy – A game that emphasizes strategic negotiation and alliance-building.
- StarCraft II – A real-time strategy game that requires quick decision-making under pressure.
- Diverse real-world personas – Scenarios that reflect everyday decision-making challenges.
Experimental Findings
Our experiments revealed that emotional perturbations have a systematic impact on the strategic choices made by the agents. However, these behaviors were often unstable and did not always align with human expectations, highlighting the complexity of integrating emotion into AI decision-making.
Conclusion and Future Work
In conclusion, while the integration of emotional sensitivity in small language model agents presents promising avenues for enhanced decision-making, challenges remain. We propose further research aimed at improving the robustness of these models against emotion-driven perturbations, which could ultimately lead to more reliable AI systems capable of nuanced human-like interactions.
As the field of AI continues to evolve, understanding the role of emotions in decision-making will be crucial for developing more sophisticated and effective interactive agents.