Specifying Computational Compliance for AI: Blueprint for a New Research Domain

AI systems, we argue, will be unable to comply with AI regulation (AIR) at the necessary speed and scale using traditional, analogue methods of compliance. Rather, compliance with these regulations can only be achieved computationally, via algorithms that run across the life cycle of the AI systems, automatically steering them toward compliance in the face of dynamic conditions. Despite their (we would argue) inevitability, the research community has yet to specify exactly how these algorithms for computational AIR compliance should behave — or how we should measure their success. To fill this gap, we specify a set of design goals for such algorithms. In addition, we specify benchmarks for quantitatively measuring whether they satisfy these design goals. By delivering this blueprint, we hope to give shape to an important but uncrystallized new domain of research — and, in doing so, incite necessary investment in it.