Regulating AI: Lessons from Scientific Computing

Motivated by the perceived importance of explainability to AI regulation, this paper examines the complex relationship between theory, as a proxy for explainability, and computation in scientific practice. Drawing from both historical and contemporary examples of scientific computing, we make three key arguments: First, we show that theory-computation relationships exist in diverse configurations, ranging from theory-foregrounded and theory-backgrounded to more nuanced arrangements where theory is selectively integrated. Second, scientific fields strategically manage the inherent opacity of some computational methods by injecting theory at critical junctures, showing that complete “explainability” is not always a prerequisite for scientific progress or utility. Third, we argue that these lessons from scientific computing have important implications for AI regulation anchored in explainable AI (XAI). We suggest that general approaches to AI oversight should be complemented with a context-specific focus on how theory and computation interact in different domains. Implications for AI regulatory regimes and Information Systems (IS) research are discussed.