The Better Part of Valor: Leveraging AI to
Mitigate Human Bias in Questions of Discretion
David
Colarusso, Suffolk
Public misapprehension regarding
the nature of current artificial intelligence (AI) has led to the adoption of
algorithmic decision aids, and in some instances decision makers, unduly
influenced by algorithmic bias—an echo of their designers’ bias. The use of
such systems in criminal justice has resulted in pushback, questioning their
fairness. Yet, proposals for combating algorithmic bias often focus exclusively
on improving an algorithm’s output. Little thought is given to the broader
context of the decision systems in which they operate, except to highlight the
danger of math-washing. Rarely is the question of relative human bias
considered in such critiques, begging the question whether such machine biases
result in more or less harm than that of the unaided human. Algorithms,
however, work to eliminate the noise found in most decisions, and this can
result in fewer, albeit different, mistakes. However, we can do better than
simply swapping biased human mistakes for less-frequent and differently biased
machine mistakes. We can use the fact that ML algorithms encode the bias of
historic data to make explicit those unjust factors driving existing decision
making, transforming the encoding of bias by an algorithm from a bug into a
feature. Drawing upon the history of burden shifting and lessons learned from
attempts to shape discretionary decisions (e.g., review committees for charging
decisions and sentencing guidelines) this paper will present a framework for
leveraging AI to mitigate human bias in questions of discretion that explicitly
considers and addresses potential legal challenges such as equal protection.
Breaking Bad: Legal Ethics and Law
Enforcement Surveillance
Tim Casey,
California Western School of Law
This
paper examines the use of evidence obtained by the government either illegally
or unethically, with specific attention to evidence based on surveillance. Deciding
whether a given action or activity is illegal turns out to be fairly
straightforward: compare the action to existing statutes and common law. Determining
whether an action is ethical turns out to be a bit more difficult. For example,
ethical rules and statutes prohibit a lawyer from violating the law, and
prohibit the use of false or misleading statements. But many law enforcement investigative
techniques depend on false or misleading statements or on violations of law by
law enforcement. Such behavior may be justified, and therefore excusable under
legal precedent, but it may nonetheless be unethical. This paper focuses on the
unethical behavior of law enforcement in electronic surveillance operations.