Artificial Intelligence for Sustainable Building Performance and Occupant Behavior: A Critical Review and a Human-Centered Framework for Hot-Climate Residential Buildings

Buildings account for roughly a third of global energy use and carbon emissions; yet, many fall short of the performance promised at the design stage. Operating buildings with less energy, lower emissions and better comfort is central to sustainable development (Sustainable Development Goals 7, 11 and 13), and AI is increasingly proposed as a tool for doing so. Occupant behavior is often blamed for the shortfall, and artificial intelligence (AI) is often offered as the cure, but the evidence linking the two is less settled than either claim suggests. This paper presents a structured narrative review of recent work on AI for building energy performance and occupant behavior, covering supervised learning, reinforcement learning, digital twins, explainable and federated approaches, and emerging language-model interfaces. I synthesize the reported savings, evidence quality and deployment barriers, and I look closely at hot-climate residential buildings, using Saudi Arabia as a reference case, where I find surveys and simulations but no field trial of occupancy-aware AI control. Three findings stand out. First, reported energy savings run from about 4% to 62%, but the spread follows the baseline and the type of evidence (field trial, calibrated twin or simulation) more than the choice of algorithm. Second, occupants matter; yet, the claim that they are the main cause of the performance gap is not well-supported. Third, privacy, interpretability and transfer between buildings, rather than model accuracy, now limit adoption. I propose a five-layer, human-centered framework that links sensing, inference, control and governance, and I set out a research agenda with explicit validation gates. The framework is conceptual and has not yet been tested in a building; I state how it can be, and I provide a reproducible search protocol for a systematic update. For sustainability, the contribution is a way to quantify and monitor claimed gains, through evidence classes and a short set of reporting indicators, and a link between the technical framework and residential energy codes, so that efficiency gains in cooling-dominated housing are credited only when they are measured.