Sentient Agent as a Judge: Evaluating Higher-Order Social Cognition in Large Language Models

TL;DR

This paper introduces SAGE, a novel evaluation framework that assesses large language models' understanding of human-like social cognition by simulating emotional and inner thoughts during interactions, providing more realistic and interpretable metrics.

Contribution

The paper presents SAGE, a new automated evaluation method that measures higher-order social cognition in LLMs through simulated emotional and mental state reasoning, validated by psychological benchmarks.

Findings

Strong correlation between emotion scores and empathy ratings

Uncovered significant gaps between top models and earlier baselines

SAGE provides a scalable, interpretable evaluation tool for social cognition

Abstract

Assessing how well a large language model (LLM) understands human, rather than merely text, remains an open challenge. To bridge the gap, we introduce Sentient Agent as a Judge (SAGE), an automated evaluation framework that measures an LLM's higher-order social cognition. SAGE instantiates a Sentient Agent that simulates human-like emotional changes and inner thoughts during interaction, providing a more realistic evaluation of the tested model in multi-turn conversations. At every turn, the agent reasons about (i) how its emotion changes, (ii) how it feels, and (iii) how it should reply, yielding a numerical emotion trajectory and interpretable inner thoughts. Experiments on 100 supportive-dialogue scenarios show that the final Sentient emotion score correlates strongly with Barrett-Lennard Relationship Inventory (BLRI) ratings and utterance-level empathy metrics, validating psychological fidelity. We also build a public Sentient Leaderboard covering 18 commercial and open-source models that uncovers substantial gaps (up to 4x) between frontier systems (GPT-4o-Latest, Gemini2.5-Pro) and earlier baselines, gaps not reflected in conventional leaderboards (e.g., Arena). SAGE thus provides a principled, scalable and interpretable tool for tracking progress toward genuinely empathetic and socially adept language agents.