Abstract
Much like electricity transformed nearly every industry it touched, artificial intelligence (AI) is becoming a general-purpose technology embedded across the fabric of modern life. There appears to be no limits on what AI can be integrated into, whether that is lifesaving systems at scale or small everyday objects. My research determines if AI can be used to create lifesaving systems and whether the sustainability impacts of AI need to be considered before beginning a project that integrates AI.
In my technical capstone, I helped build an interactive firearm safety education system for young children ages 4 to 7 using a social robot, augmented reality (AR) and AI. It was determined that a system like this could help scale current firearm safety training efforts that use Behavioral Skills Training (BST), an active learning strategy that has been proven to teach skills to young children but requires special instructor training. To follow BST, the system used a NAO social robot to provide engaging instruction and an AR simulation to allow the children to act out the steps in a safe, realistic looking environment. While the use of AI is justified in this project because it has the potential to save countless young lives, many projects lack a clear basis for evaluating whether their benefits outweigh the environmental costs.
In my STS research, I examined the tension between AI’s growing environmental footprint and its potential to drive meaningful positive change because while it is well known that AI systems consume significant amounts of energy, water, and materials, there is no practical framework for determining whether those environmental costs are justified by an AI tool’s benefits. I tried to fill this gap by developing an analytical framework, grounded in Actor Network Theory (ANT), that analyzes different AI tools across three levels based on the scale and nature of their benefits, and then evaluates each level against its environmental costs. I used ANT because this problem requires a systems approach that analyzes the impacts of network interactions between developers, corporations, users, government, and elements of the natural environment. I found that the scale of models and their impacts matter significantly, highlighting that the benefits of larger, resource-intensive models do not necessarily scale in proportion to their environmental costs.
Considered together, my capstone project and STS research demonstrate both the potential value of AI systems and the need for a structured way to evaluate when their use is justified. Drawing on ANT, this emphasizes that such frameworks must account for the interconnected roles of developers, users, institutions, and environmental systems in shaping AI’s overall impact.