Social Reasoning
Overview
Social reasoning is the ability to interpret interactions by inferring what people know, believe, intend, feel, and expect, and how norms or relationships shape their actions. AI systems are evaluated with theory-of-mind stories, dialogue and long-thread questions, multimodal video benchmarks, and multi-agent games with measurable outcomes. Approaches include explicit belief and relationship models, evidence-grounded prompting, supervised fine-tuning, reinforcement learning, self-play, and norm-learning modules. Models can do well on familiar or structured tasks but remain brittle when cues are uncertain, multimodal, culturally dependent, or counterfactual, leaving open whether plausible answers reflect robust, adaptive understanding.
Latest papers
41 papers5 papers in the last four weeks, down 17% on the four weeks before. 0.0% of all new papers.