The AI industry is of interest to policymakers and dangerously ahead of current policy attempts for three key reasons: its simultaneous similarity to and utter distinctness from the familiar software industry, the speed at which it’ s been adapted, and the naturally evolving supply chain that supports its implementation. Current efforts to produce policy ensuring the safe use of ML-powered tools largely focus on understanding the technology’ s unique properties and risks. But this work, lethargic relative to the AI industry’ s superhuman speed, does not consider how the real-world structures implementing ML-powered tools might interact with their inherent risks. This projects seeks to map and visualize a complex, interdependent AI supply chain, establish some of the unconsidered risks that could arise from the current deployment structure, and experimentally show some of these risks in an industry-like setting.

I sought out SuperUROP to get experience with the defined deliverables of a substantial, long term research project. My interest in safe and explainable ML models sets me up well to consider the regulatory implications of failures in the field, and I’ m looking forward to understanding AI use outside of a research context. The SuperUROP framework allows me to both research what I love and develop my communication skills in a scientific context.