Algorithmic Transparency for Governments
Governments have increasingly been using predictive algorithms created through the use of machine learning to assist them in making a wide variety of decisions. They consult predictive algorithms about everything from promoting teachers to assigning police patrols, paroling prisoners, and investigating child welfare. When the governments in question are democratic, there must be some way for citizens to hold them accountable for their use of these algorithms, and to consider whether the algorithms are inaccurate, biased, or otherwise flawed. Yet that kind of accountability requires an understanding of how such algorithms are constructed, and requires disclosure by governments of the choices made and the tests performed during such construction.
This talk will walk through the process of constructing a predictive algorithm, and consider the value choices made in doing so. It will then present the results of a study that filed 42 open records requests in 23 states to seek information about six predictive algorithm programs. That study concludes that most governments do not currently disclose enough information about predictive algorithms to enable evaluation of their accuracy and fairness. Publicly-deployed algorithms will be sufficiently transparent only if (1) governments generate appropriate records about their objectives for algorithmic processes and subsequent implementation and validation; (2) government contractors reveal to public agencies sufficient information about how they developed algorithms; and (3) public agencies and courts treat trade secrecy claims as the limited exception to public disclosure that the law requires.
Robert Brauneis is Professor of Law and Co-Director of the Intellectual Property Law Program at the George Washington University Law School. He is a founding member of the Project Board of the Munich Intellectual Property Law Center. He is also a past President of the Giles Rich Inn of Court and a Trustee of the Copyright Society of the USA, and is an Advisor to the Restatement of the Law, Copyright.