Dialogues with algorithms

TL;DR

This paper discusses the importance and challenges of enabling dialogues between humans and rule-based legal software, emphasizing the need for transparency and contestability in law enforcement algorithms.

Contribution

It highlights the potential for formal methods to facilitate dialogues with legal algorithms and examines the trade-offs involved.

Findings

Dialogues with legal algorithms are highly desirable for transparency.

There is a tension between dialogue expressibility and computational feasibility.

Formal methods can potentially enable such dialogues in the future.

Abstract

In this short paper we focus on human in the loop for rule-based software used for law enforcement. For example, one can think of software that computes fines like tachograph software, software that prepares evidence like DNA sequencing software or social profiling software to patrol in high-risk zones, among others. An important difference between a legal human agent and a software application lies in possible dialogues. A human agent can be interrogated to motivate her decisions. Often such dialogues with software are at the best extremely hard but mostly impossible. We observe that the absence of a dialogue can sincerely violate civil rights and legal principles like, for example, Transparency or Contestability. Thus, possible dialogues with legal algorithms are at the least highly desirable. Futuristic as this may sound, we observe that in various realms of formal methods, such dialogues are easily obtainable. However, this triggers the usual tension between the expressibility of the dialogue language and the feasibility of the corresponding computations.