Trust But Verify: A Guide to Algorithms and the Law
65 Pages Posted: 27 Apr 2017 Last revised: 27 Apr 2018
Date Written: April 27, 2017
The call for algorithmic transparency as a way to manage the power of new data-driven decision-making techniques misunderstands the nature of the processes at issue and underlying technology. Part of the problem is that the term, algorithm, is broad. It encompasses disparate concepts even in mathematics and computer science. Matters worsen in law and policy. Law is driven by a linear, almost Newtonian, view of cause and effect where inputs and defined process lead to clear outputs. In that world, a call for transparency has the potential to work. The reality is quite different. Real computer systems use vast data sets not amenable to disclosure. The rules used to make decisions are often inferred from these data and cannot be readily explained or understood. And at a deep and mathematically provable level, certain things, including the exact behavior of an algorithm, can sometimes not be tested or analyzed. From a technical perspective, current attempts to expose algorithms to the sun will fail to deliver critics’ desired results and may create the illusion of clarity in cases where clarity is not possible.
At a high-level, the recent calls for algorithmic transparency follow a pattern that this paper seeks to correct. Policy makers and technologists often talk past each other about the realities of technology and the demands of policy. Policy makers may identify good concerns but offer solutions that misunderstand technology. This misunderstanding can lead to calls for regulation that make little to no sense to technologists. Technologists often see systems as neutral tools, with uses to be governed only when systems interact with the real world. Both sides think the other simply “does not get it,” and important problems receive little attention from either group. By setting out the core concerns over the use of algorithms, offering a primer on the nature of algorithms, and a guide on the way in which computer scientists deal with the inherent limits of their field, this paper shows that there are coherent ways to manage algorithms and the law.
Keywords: algorithms, governance, accountability, internet, cyber, technology, bias, discrimination, computational methods, law, big data, computer science, code, transparency, machine learning
JEL Classification: C6, K00, K3, K4, K1
Suggested Citation: Suggested Citation