Artificial Intelligence (AI) is increasingly integrated into Building Automation Systems (BAS) to enhance energy efficiency and occupant comfort. Yet, rather than functioning as neutral optimization tools, AI in BAS operates within fragile infrastructures, limited resources, and institutional politics. We present a qualitative study of 23 interviews with energy professionals, AI researchers, and student representatives at the University of Toronto, an institution recognized for its sustainability leadership. Participants expressed ambivalence: AI was valued for forecasting and optimization, yet concerns arose around legitimacy, labor demands, and environmental paradoxes. Fairness in occupant comfort was highlighted, not as an inherent property of models but as a situated practice shaped by infrastructural governance negotiated across roles and inequities. Communication also emerged as a form of occupant agency, where human, machine, and AI-mediated dialogue makes automated decisions legible and contestable. These findings reframe AI in BAS as socio-technical infrastructure and inform our design recommendations for transparent, participatory, and just systems.
ACM CHI Conference on Human Factors in Computing Systems