From Passive to Persuasive: Steering Emotional Nuance in Human-AI Negotiation

Researchers developed a new method called activation engineering to make AI language models express more human-like emotions in conversations. The technique uses targeted interventions on LLaMA 3.1-8B to enhance emotional characteristics like positive sentiment and personal engagement without extensive fine-tuning.

• →Activation engineering can steer LLaMA 3.1-8B to exhibit more human-like emotional nuances without extensive fine-tuning.
• →Attribution patching was used to identify causally influential components for emotional expression in AI models.
• →Emotional expression vectors derived from contrastive text pairs significantly enhanced conversational characteristics.
• →Steered AI responses showed increased positive sentiment and more frequent first-person pronoun usage.
• →The research provides a precise and interpretable framework for improving conversational AI systems.