A Behavioral Finance Framework for Balancing AI Accuracy and Operational Carbon Emissions

The increasing computational demands and related carbon emissions of Artificial Intelligence (AI) models necessitate careful consideration of their environmental impact. This concern is especially important when these models are trained in regions with energy shortages or carbon-intensive power grids (such as coal, natural gas, or oil), as their operational footprint can worsen environmental strain. To address this challenge, we develop a behavior-driven computational framework for selecting AI models that balance accuracy and environmental sustainability. To capture real-world decision-making under sustainability constraints, we propose a behavior-driven framework that analyzes tradeoffs between energy use, carbon emissions, and model performance using insights from Behavioral Portfolio Theory and Cumulative Prospect Theory. We quantify emissions from energy use during supervised fine-tuning across multiple model architectures and datasets, incorporating these results within a multi-objective decision model. Our framework provides actionable insights for selecting AI models that minimize environmental impact without sacrificing accuracy. By prioritizing models that balance performance and sustainability, we provide AI developers with practical tools to lower the carbon footprint of AI technologies.