Abstract

Functionalist Perspective on Emotions in AI: A Review of Roles, Mechanisms and Impacts TREO Talk Paper Eun Hee Park Old Dominion University epark@odu.edu Mala Kaul University of Nevada, Reno mkaul@unr.edu Chad Anderson Miami University, Ohio ander556@miamioh.edu Abstract Recent advances in artificial intelligence (AI) make it increasingly important to understand and theorize how emotion-related processes shape system behavior and decisions. Prior research such as in affective computing and human-computer interaction has focused largely on AI emotion detection and expression, but less attention has been paid to how these processes function within AI systems and shape their decisions. Recent interpretability research by Anthropic has demonstrated that large language models encode internal representations of emotion-related concepts that systematically influence their output and behavior. These findings provide empirical evidence that emotion-like mechanisms can play a functional role in shaping system behavior while highlighting the need for clearer ways to understand how such mechanisms operate, particularly given the opacity of modern AI systems. Motivated by this gap, this research examines “emotion in AI” through a conceptual literature review grounded in a functionalist perspective, that views emotions in terms of their roles in evaluation, regulation, and action. Rather than assuming that AI systems explicitly implement emotion modules, we focus on what these processes do and how their effects can be observed in system behavior. Based on this perspective, we develop a classification scheme (Figure 1) organizing prior research around eight functional roles: appraisal, representation, generation, prioritization, action readiness, adaptation, coordination, and feedback. This scheme provides a structured way for interpreting how internal processes shape AI cognition and decision-making. Understanding these mechanisms and making sense of the emerging evidence on internal representations, is critical for improving the safety, reliability, and quality of human-AI interaction. Figure 1. Functional Roles of Emotion in AI: A Classification Scheme For this review, we conducted keyword-based searches in major databases, including AIS Basket journals and Web of Science, using terms such as emotion, affective computing, and emotional AI. The dataset spans information systems, AI, robotics, and human-computer interaction. Currently, in the data analysis phase, we are classifying studies by proposed functional roles to identify patterns across themes and constructs. This study contributes in three ways. First, we develop a theoretically grounded structure for organizing a fragmented body of research on functional emotion in AI. Second, we provide an interpretive lens linking observable behaviors to emotion-related function within AI systems. Third, we identify underexplored areas, such as control mechanisms and feedback processes, critical for understanding user-interaction related decisions made by AI systems. Practically, this suggests that even when AI systems are not fully transparent, their functional effects can be considered in their design and governance, to ensure safe and reliable human-AI interaction. Overall, this work reframes emotion in AI as a functional aspect of system behavior and develops a future research agenda for decision-making, trust, and responsible AI design.

Download

Share

COinS