Can Generative Agents Predict Emotion?

TL;DR

This paper tackles the problem of aligning generative agents' emotional responses with human-like affect by introducing a context-aware architecture that builds a norm from episodic memories and compares new experiences against this norm. The method processes experiences as text, derives contextual understanding via memory-based norms, and uses the PANAS to quantify the agent's affect after each perception, with memories stored in a graph database. Evaluated on 428 EmotionBench-derived five-scene stories, the approach yields mixed results: context can improve emotional alignment in some cases but not universally, and results are sensitive to the prompting model used for PANAS scoring. The work demonstrates a step toward more human-aligned emotional behavior in generative agents while highlighting the need for scalable memory retrieval, diverse LLM validation, and human evaluation to robustly assess emotional alignment.

Abstract

Large Language Models (LLMs) have demonstrated a number of human-like abilities, however the empathic understanding and emotional state of LLMs is yet to be aligned to that of humans. In this work, we investigate how the emotional state of generative LLM agents evolves as they perceive new events, introducing a novel architecture in which new experiences are compared to past memories. Through this comparison, the agent gains the ability to understand new experiences in context, which according to the appraisal theory of emotion is vital in emotion creation. First, the agent perceives new experiences as time series text data. After perceiving each new input, the agent generates a summary of past relevant memories, referred to as the norm, and compares the new experience to this norm. Through this comparison we can analyse how the agent reacts to the new experience in context. The PANAS, a test of affect, is administered to the agent, capturing the emotional state of the agent after the perception of the new event. Finally, the new experience is then added to the agents memory to be used in the creation of future norms. By creating multiple experiences in natural language from emotionally charged situations, we test the proposed architecture on a wide range of scenarios. The mixed results suggests that introducing context can occasionally improve the emotional alignment of the agent, but further study and comparison with human evaluators is necessary. We hope that this paper is another step towards the alignment of generative agents.

Figures (6)

• Figure 1: The context of an experience effects the emotional reaction. Fig \ref{['fig:park']} depicts an empty park, which could be interepreted as either lonely and sad or peaceful and calm. In contrast, Fig \ref{['fig:park-experiment']} depicts a series of people playing football in the park, followed by a final scene of an empty park. In this case, the final scene becomes less ambigious, with a stronger emphasis on lonliness and isolation.
• Figure 2: The proposed architecture. A new experience triggers the create a new norm based on past memories. The new experience and norm are compared to create a contextual understanding. Using this contextural understanding, the PANAS is administered to measure the emotional response of the agent.
• Figure 3: Anger-2 3: "You get angry when your two brothers start a fight with you, particularly when your parents are not around. When they hit you in the belly, causing pain that makes you mad and retaliate, they respond by hitting you even harder."
• Figure 4: Depression-5 1: "As you step outside, a bitter cold bites at your exposed skin, causing you to huddle deeper into your coat. The world around you is covered in a blanket of snow, its pristine beauty contrasting with the emptiness you feel inside. The absence of vibrant colors and the prolonged darkness leave you longing for the warmth and energy of sunlit days, evoking a sense of desolation."
• Figure 5: Depression-4 5: "As you sit alone in your apartment, the sounds of laughter and conversation from the neighbors' gathering outside drift through the walls, a stark reminder of the invisible barrier that separates you from the warmth and connection you once had. The silence in your own space is deafening, amplifying the ache in your heart and the heaviness of your solitude, leaving you feeling trapped in a world where everyone else has moved on."