Meet the Book Author- AI and the Rule of Law: The Necessary Evolution of a Concept

Paul Burgees, Monash University

n our Meet the Book Author Series, the Journal of Law and Society and the Centre of Law and Society provide first-hand accounts from authors who have recently contributed notable socio-legal books to their respective fields. In this post, we hear from Paul Burgees from Monash University, whose new book AI and the Rule of Law: The Necessary Evolution of a Concept was published in 2024 with Bloomsbury Publishing.

This is a book that answers a question that has puzzled me for quite some time: how will our concept of the Rule of Law need to change in order to account for the exercise of power by Artificial Intelligence (‘AI’) in the future?

What is important to say is that I am not an AI researcher in the technological sense. For this reason, the book is written from the perspective (and is largely for those interested in) constitutional theory, public law, and legal theory.

Two fundamental premises underlie the book’s thesis. The first is that the use of AIs in the future is likely to increase. More particularly, as a result of the potential efficiencies offered by AIs, the use of AI by states in their exercise of power seems likely to both increase (in general) and to become more and more sophisticated. This raises the potential of power being exercised by an algorithm in ways that may be subject to ever decreasing levels of human control or supervision.

The second and third premises are fundamentally interconnected. The second is that the Rule of Law is a concept that remains valued and valuable. The third is that the Rule of Law is able to change its conceptual boundaries. The Rule of Law is clearly of continued relevance worldwide as many states claim adherence to it – notwithstanding its contested nature – perhaps as the idea of a concept that, in the broadest and most acceptable terms, opposes the arbitrary application of power cannot be a bad thing.

In bringing these premises together with the question I pose, I wanted to shine some sort of light on the potential problems that exist when we try to proclaim continued adherence to a fundamental concept that is not only contested in its scope and definition, but also that has been conceived in very different ways over time. And, what’s more, to recognise the potential problems that arise where power is exercised in different ways in the future.

In order to do this, I adopted an approach that was unconventional in terms of Rule of Law ideas and definitions. The use of the thin/thick or formal/substantive categories is relatively commonplace in the literature that considers the ways in which the concept can be defined. Given the temporal nature of the question, and as there seemed to be an immediate temporal distinction between the forms of ideas, I adopted a categorisation of a Past Rule of Law and a Present Rule of Law. 

The dual conceptual categories each included Rule of Law conceptions that are frequently invoked to evidence what the Rule of Law is. For example, the Past Rule of Law was comprised by the accounts of: Aristotle, Locke, Dicey, Hayek, Fuller, and Raz. The Present Rule of Law is comprised by more recent Rule of Law accounts provided by authors or institutions like Jeff King, Tom Bingham, Martin Krygier, and the United Nations. 

In order to identify the relative problems and protections that either arose or subsisted in terms of each category, I then explored three hypothetical ways in which AI may exercise state power in the future: AI making administrative decisions; AI creating secondary legislation; and, AI creating secondary legislation. These move, respectively, from extant uses of the technology toward the most extreme hypothetical that would likely never eventuate even if the technology did exist: replacing the current human-staffed legislature with a virtual legislature staffed only by elected algorithms.

As these ideas extend into extreme hypothetical territory, in order to avoid extended discussions of Terminator-esque / AI doomsday outcomes, I couched my consideration of the problems and protections in both of the Past or Present Rule of Law paradigms in a non-dystopian context. In other words, I did not consider the possibility that allowing AI to operate in any of the hypothetical ways would result in the end of the world; although, I also did not consider AI’s use in these ways to be a global panacea. 

The result of this was a somewhat oxymoronic outcome that was both novel yet unsurprising. The use of AI in the exercise of power in the ways hypothesised resulted in a number of problems that could be seen as being familiar Rule of Law issues. These included issues relating to predictability, accountability, and speed; yet, the nature of these problems differed suitably from the versions of the problems that were being responded to in the accounts that are included within the Past and Present Rule of Law. 

As one illustration, the problem of speed in the Past Rule of Law can be summed up by paraphrasing a frequently deployed idea of Aristotle: there should be some form of due process when legislation is created. This stems from the idea that humans – as Aristotle was talking about human legislators – make errors, resulting in bad laws, when making decisions quickly. This does not, however, occur with algorithms – as algorithms are able to make decisions exponentially faster than humans without error. 

This does itself lead to a different problem (particularly when considered in terms of the hypothetical in which AIs make primary legislation): all of the elected AIs could be involved in the proposal, extended debate, amendment, and passage of a piece of legislation in a fraction of a second. This raises a range of speed-related problems in terms of the Rule of Law, but these are fundamentally different in their nature to the often-cited problem of speed.

In consequence of the different forms of problems that arise (even superficially similar problems), I conclude that conceptions that make up neither the Past or the Present categories of the Rule of Law are not capable of providing the sort of protections from arbitrary power that we expect from the concept. And, with that in mind, I close the book with suggestions as to how the Future Rule of Law may be reconceived to continue to provide protections.