The Necessity of Revisiting the Rule of Law and the Rise of Algorithmic Decision Making - Revisiting the Rule of Law book forum
30.11.2023
I was not convinced that a small (66 page) book that – in addition to being written for a principal audience of students and teachers of legal and political thought – was written for ‘anyone who wants to understand more about how we think, speak, and write about the rule of law’ (at 3) could provide a creative approach or motivate substantive further academic ideas about the concept. I was wrong.
Across only three logically (yet innovatively structured) short parts, Kristen Rundle stimulates and provokes engagement with the concept of the Rule of Law in Revisiting the Rule of Law. The three parts – that first examine the methodology associated with theorising the Rule of Law before considering the entangled way in which the concept is frequently presented – culminate in a part that explores ‘what may be missing or under-examined within theoretical engagement’ (at 2) and identifies a need to revisit the Rule of Law.
In doing this, in identifying the importance of the subjects and the officials of the Rule of Law (at 53-59), Rundle identifies issues that will, I believe, form the foundation of the Rule of Law’s theoretical discourse over the coming decades: the rise of technology and algorithmic decision-making in the exercise of state power. The issues raised in these respects whilst clearly stated were also rich and beguilingly complex. In the brief paragraphs below, I set out not only why I think that the issues that were set out by Rundle are so important, but also why their being raised in the way that they are raised in the book is, now, more important than ever.
In recent years, I have become increasingly interested in the intersection of artificial intelligence, algorithmic decision-making, and the Rule of Law. In my forthcoming book – AI and the Rule of Law; The Necessary Evolution of a Concept (Hart, 2024) – I pose the question: how will the concept of the Rule of Law need to evolve in order to account for artificial intelligence being used to exercise public power in the future? In seeking to answer this question, I do not assume that the standard paradigm that has traditionally been adopted in relation to the Rule of Law will or should continue to operate. In short, I seek to identify the different ways in which we will need to conceive of the Rule of Law in the future. With this in mind, with an eye to Rundle’s project, it is likely of no surprise that I am a clear advocate of the need to revisit our conceptions of the Rule of Law.
In considering the need for a new conception of the Rule of Law, it is useful to briefly illustrate ways in which the concept of the Rule of Law and use of AI in the exercise of public power may interrelate. The first represents the use of AI to assist in visa processing. The vast number of visa decisions that are required to be made means the development of AI systems to assist human decision-makers would represent a key efficiency. The second represents the use of AI exercising delegated power to make secondary legislation. For example, in amending a schedule of proscribed substances where an AI system could operate like an expert body that lays a recommendation before the legislature in order to change the schedule. As a more extreme example, the legislature could be replaced by elected algorithms as the sole creator of primary legislation.
These examples each reflect an exercise of public power at varying levels of near-term application. (These examples are explored in detail, each in a separate chapter, in my book.) The third is clearly a thought experiment and does not represent the technical reality of AI at present. The second example represents a conceivable application in the near future. The first example is already a reality in some jurisdictions.
These examples illustrate the technologies’ overlap with the Rule of Law. As a concept that relates to the arbitrary exercise of power, the exercise of power through these different means could represent a different form of the exericise of power – particularily in relation to the people of the Rule of Law that Rundle explores (at 53). In the first example, where the AI system is merely a tool being used by the human decision-makers, it may be said there is no need to revisit or change our view of the concept. It is less clear that popular ideas of the Rule of Law may apply to the artificial decision-makers in the other examples. After all, all previous ideas of the Rule of Law were conceived to constrain the exercise of power by humans. If decisions are being made by non-humans, it is relevant to ask whether the concept can still provide protections and, if it does not, how would the concept need to be concived in order to continue to do so. It is this thought that frames the question I seek to answer in my book.
The research question that frames my book does not precisely mirror Rundle’s project; but there is a similarity in motivation. This can most clearly be seen when the Rundle’s third part is considered: revisiting the Rule of Law needs ‘to be an exercise in noticing what remains under-examined or underdeveloped within theoretical treatment of the idea, and asking why’ (at 48). My early work involved considering the historical context of the Rule of Law (including the importance of concept’s history) and the nature and boundaries of the idea. In tandem with this I developed a parallel strand of research based on my interest in artificial intelligence (‘AI’) relating both to AI’s legal personhood and the augmentation of democracy. In thinking of these two ideas it became apparent that the increasing and frequent use of AI by governments in the exercise of state power was, to adopt Rundle’s phrasing, ‘under-examined or under-developed’. Rundle’s focus on people (at 53) makes it even clearer that consideration of what may constitute a subject of the Rule of Law is now more important than ever.
This remained (and remains) the case despite what seemed to be a clear and obvious connection to the core idea of the Rule of Law: the opposition to the arbitrary exercise of power by the state. (An example of the prevalence of AI use in exercising power is provided by a recent NSW Ombudsman’s report.) Where the benefits of increased speed and efficiency that AI can bring to state-based decision making are clear, the use of AI by the state seems only likely to increase. Where this is the case, it seemed important to consider whether the concept of the Rule of Law – as popularly conceived (in the least) – can continue to provide protection against artificially wielded exercisers of power. Yet, this is not something that is being widely considered. (One notable example, is the recent work of Zalnieriute, Bennett Moses, and Williams.)
In other words, the motivating factor behind my recent work has been the realisation that there are ideas of the Rule of Law that – despite the now decades-old and ever-expanding literature relating to the meaning and concept of the Rule of Law – are both ‘under-examined and underdeveloped’ when it comes to the different ways in which power may be exercised. A simple example may illustrate this. The Rule of Law as it is traditionally and commonly conceived requires a form of predictability or procedural certainty in relation to the process of law-making. (I previously identified these two elements as motivating factors that exist behind the most frequently relied upon Rule of Law ideas.)
This will mean that if a state wishes to adhere to any idea of the Rule of Law (Past or Present), the AI that it uses in ways that reflect the examples provided earlier must be explainable. Yet, explainability is not something that AI (at least in its present form) is very good at. It is often the case that the way in which a decision is made by an algorithm is unknown; the decision is made within a black-box. This raises fundamental questions regarding whether the use of AI by the state in the exercise of power – where a decision-making process cannot be known – can satisfy the Rule of Law.
Another way in which there may be an ‘under-examined and underdeveloped’ aspect of the application of AI in the exercise of power in relation to the people of the Rule of Law may relate to the speed at which power is exercised. One of the ideas of the Rule of Law – frequently associated with Aristotle in Rhetoric – is that the process of law-making should not be one that is subject to emotion and is not one that should be conducted hastily. The introduction of AI to the process may facilitate an exponential increase in the speed of decisions. Yet, whilst this may be initially thought to represent a problem for the Aristotelean idea, closer examination and development of the issue reveals this may actually not relate to that obvious connection.
Aristotle’s point seems to relate to the issues associated with the introduction of human emotion into the process. Where a human makes a decision rapidly and rashly, this will not result in good laws. But an AI’s increased speed of decision making does not result in an increase in errors and it does not reflect human emotion.
This means that the problem as conceived by Aristotle has no purchase; but there is still a different problem. This different speed problem is one that reflects an external manifestation. Aristotle’s problem could be said to be internal; it relates to the way in which the process of law-making is impacted by speed, as the laws that result are not as well conceived. The problem associated with the use of AI is an external problem; the speed at which the laws are made impacts the subjects of the law, as the increased speed means it is harder to know or understand the laws that should otherwise guide their actions. When considered in this way, it becomes apparent that the use of AI and other technologies in the exercise of power by the state raises new and different problems for the concept of the Rule of Law – problems that it was not necessarily intended to address.
It was in these respects that Rundle’s motivating aim regarding the ‘under-examined and underdeveloped’ ideas of the Rule of Law in terms of the people of the Rule of Law struck a chord. The use of AI in the exercise of public power was unconceived and inconceivable by any of the theorists that we commonly rely on when we seek to identify the meaning and content of the concept. This raises fundamental questions about what the Rule of Law will need to be in the future if it is to continue to be the concept that we rely on to protect us from the arbitrary application of power by the state.
Any new conception of the Rule of Law must, in addition to providing a defence against extant forms of the exercise of arbitrary power, must also respond to the problems that the exercise of power by AI raises: the problems most pertinent to solve relate to explainability, speed, and accountability. For example, using speed as an example, the reconceived idea of the Rule of Law would need to temper the exercise of power both in a way that must be made slowly enough to be externally comprehended and in a way that ensures the exercise of power can be communicated in a way that can be comprehended by those that are subject to the power.
In identifying the future idea of the Rule of Law, we need to consider the specific problems that AI’s exercise of power raises and that we should do this from a position that does not rely on the previous canon of the Rule of Law. This is because each of those authors provide a solution to a problem that they have in their societies – their highly contextualised accounts cannot (without anachronism) be reconceived in a way that responds to the challenges posed by AI. It is the concept itself that must be reconceived.
Whilst the invention of new ideas may not sit well with a modern academic practice that requires recourse to authority or predent, the ‘creation’ of a concept Rule of Law from a blank canvas is not unprecedented. For example, Dicey described a concept without recourse to any preceding accounts of (what we now consider) the Rule of Law. In this respect, his account was largely a description of society at the time (or up until that point).
One way to reconceive the concept would be to adopt a form of teleological approach to the Rule of Law advocated by Martin Krygier. By conceiving the Rule of Law to be a solution concept that responds to the problem of the exercise of arbitrary power (whatever form that may take), the concept could respond to the exercise of arbitrary power by human and non-human actors without the conceptual baggage associated with accounts that focus on more specific problems. (I have previously argued against this approach – but my arguments fall away where view reliance on prior Rule of Law ideas is abandoned.) Whilst this does not prescribe a core ‘laundry list’ of desiderata, it provides the flexibility that is necessary to ground any beneficial reconception.
In this respect, the exercise of power by AI as a non-human entity within the state falls squarely into the category that Rundle uses to motivate our need to revisit the Rule of Law. This need to reconceive the concept of the Rule of Law reflects not only the title of Rundle’s book, but also the necessity of doing so. For this reason, Rundle’s consideration is not only timely but essential. .
Paul Burgess is interested in all things related to the concept of the Rule of Law.
Suggested citation: Paul Burgess, ‘The Necessity of Revisiting the Rule of Law and the Rise of Algorithmic Decision Making - Revisiting the Rule of Law book forum’ (30 November 2023) <https://www.auspublaw.org/blog/2023/11/the-necessity-of-revisiting-the-rule-of-law-and-the-rise-of-algorithmic-decision-making-revisiting-the-rule-of-law-book-forum>