Title: Accuracy Bounding: A Regulatory Path Forward for the Algorithmic Society

Abstract: Use of algorithms in a widespread array of decision making and automation tasks is increasing even as regulators and the general public express unease regarding the potential for negative social effects from the use of these technologies. Three prominent and domain specific concerns continue to resurface in media and popular discourse as an algorithmic society emerges: bias, surveillance, and economic disruption. This paper recommends a straightforward but novel proposal for addressing these prominent ills of algorithms in the short-term and for inspiring new forms of solutions even in longer-term regulation: accuracy bounding. Accuracy bounding would limit the performance features of algorithmic products along just one metric of interest, accuracy, thereby creating a regulatory and economic environment that explicitly incentivizes algorithmic product improvement for other quality indicators, such as robustness, fairness, and transparency. This proposed regulatory method would offer a configurable and transparently specifiable means of addressing algorithmic ills in the short term as the legal and technical communities are given time to develop shared practical definitions of quality as applied to enhance AI performance and accountability.

Bio: Aileen Nielsen is a Fellow in Law & Tech at the Center for Law and Economics at ETH Zurich.