The Gaps that Never Were: Reconsidering Responsible AI's Principle-Practice Problem

Metaphors shape perception and action. Across domains of responsible AI, the gap-bridge' metaphor proliferates. Troubled by an ostensible gap between principles and practices, the field has produced an array of bridging instruments in the form of toolkits, guidelines, and frameworks. The corpus of these instruments is expansive and robust. Yet, the principle-practice problem persists. Rather than propose new and better bridging devices, we step back to interrogate the metaphor itself. Reading metaphors as artifacts, we breakdown the elements of bridges and gaps, abstracting these into identifiable characteristics: separation, linearity, and stasis. We then draw on fieldwork with three startup organizations to understand whether and how gap-bridge characteristics manifest. The organizations under study all have AI in their portfolios and are actively working to (re)construct their principles and practices, thus showcasing the principle-practice problem in action. Findings show the gap-bridge metaphor misfits with these organizations' in-situ efforts, in which principles and practices are integrated, nonlinear, and subject to dynamic values that transform across time and circumstance. With these findings in mind, we consider alternative metaphors and invite collective reconsideration of principles, practices, and their relationship to responsible AI.