Year: 2026
Type: conference-paper
Abstract: Affective computing, the development of systems that recognize, interpret, and simulate human emotions, has advanced rapidly through deep learning and multimodal fusion techniques.Yet this technological progress has significantly outpaced foundational understanding of what emotions are, how they function socially, and what it means for machines to simulate them.This position paper argues that current affective systems are critically flawed because they rely on correlational patterns rather than established psychological theories of affect, are trained on biased data that systematically fails to generalize across demographic groups, and operate within inadequate ethical and regulatory frameworks that cannot protect emotional privacy or prevent harm.Drawing on empirical work in affective science, computational fairness research, and philosophical accounts of emotional expression, we argue for a theory-first approach that integrates psychological models, mandates rigorous fairness auditing across intersectional demographics, treats affective data as a protected category requiring heightened safeguards, and recognizes fundamental limits to emotion recognition systems.Without such grounding, affective computing risks systematically encoding bias, enabling emotional manipulation, and eroding authentic human connections that define meaningful social experience.
Author: Wajdi Zaghouani
Institution: Northwestern University in Qatar
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
Funder: Qatar National Research Fund
OpenAlex ID: W7166528493