The Rule of Law After Artificial Intelligence

A New Volume Asks Whether Legal Systems Designed for Human Actors Can Survive the Arrival of Non-Human Decision-Makers

There is a particular kind of silence in a courtroom. It has weight. The architecture of justice, from the raised bench to the witness stand, was designed to produce a specific phenomenology: the feeling that something consequential is happening, that human judgment is being exercised in a space set apart from ordinary life. Walk into any European court and notice how the room itself argues for the legitimacy of what occurs within it.

Now imagine that same room, but the decision has already been made. Not by the judge, but by a system that processed the case file overnight, weighed the precedents, calculated the risk scores, and produced a recommendation that arrives on the bench as a fait accompli. The judge can override it, technically. But the architecture of the room no longer matches the architecture of the decision.

This is the territory that The Rule of Law After Artificial Intelligence, a new paperback volume available through the University of Chicago Press, attempts to map. The book arrives at a moment when European policymakers are grappling with precisely this disjunction: legal frameworks built for human agency encountering systems that operate according to different logics entirely.

What the Rule of Law Actually Requires

The phrase "rule of law" gets invoked so frequently in AI governance debates that it risks becoming decorative, a rhetorical flourish rather than a substantive constraint. But the concept has specific content. It means that power must be exercised according to publicly known rules, that those rules must be applied consistently, that decisions must be explainable and contestable, and that no one, including the state, stands above the law.

Each of these requirements becomes complicated when algorithmic systems enter the picture.

Consider consistency. A human judge might decide similar cases differently based on factors that are difficult to articulate but nonetheless legitimate: the demeanor of a witness, the sense that something doesn't add up, the accumulated wisdom of years on the bench. An algorithmic system, by contrast, will decide similar cases identically, which sounds like an improvement until you realize that the similarity is defined by whatever features the system was trained to recognize. Two cases that look identical to the algorithm might be meaningfully different in ways the system cannot perceive.

Or consider contestability. The rule of law requires that decisions can be challenged, that reasons must be given, that affected parties can argue their case. But what does it mean to challenge a decision when the decision-maker is a neural network with millions of parameters? The system cannot be cross-examined. It cannot explain its reasoning in terms that map onto legal categories. It can only produce outputs.

The European Context

This is not an abstract problem for European governance. The EU AI Act, which entered into force in 2024 and whose provisions are now being implemented across member states, explicitly addresses AI systems used in the administration of justice. Such systems are classified as high-risk, subject to requirements for transparency, human oversight, and documentation.

But the Act's requirements, however carefully drafted, operate at a different level than the questions this volume raises. Compliance with the AI Act is a matter of technical standards, conformity assessments, and regulatory oversight. The rule of law is something else: a set of normative commitments about the relationship between power and accountability that cannot be reduced to a checklist.

Research from Monash University has explored how AI systems interact with rule of law principles, noting that the challenge is not simply technical but conceptual. Legal systems evolved to regulate human behavior through mechanisms that assume human capacities: the ability to understand rules, to be deterred by sanctions, to feel the weight of moral obligation. Algorithmic systems have none of these capacities. They optimize for objectives. They do not understand anything.

The Phenomenology of Algorithmic Judgment

What does it feel like to be judged by a machine? This is not a question that appears in most policy documents, but it matters. The legitimacy of legal systems depends not only on their formal properties but on their experiential qualities: the sense that one has been heard, that the decision-maker engaged with the particulars of one's situation, that justice was not merely done but was seen to be done.

Algorithmic systems disrupt this phenomenology. Even when they produce accurate outcomes, even when they reduce bias compared to human decision-makers, they alter the experience of being subject to legal authority. The person whose case was decided by an algorithm may have received a fair outcome, but they did not receive the experience of being judged by another human being who looked them in the eye and rendered a verdict.

This is not sentimentality. It is a recognition that legal systems serve functions beyond the efficient resolution of disputes. They perform the social drama of accountability. They make visible the exercise of power. They create moments when the state must justify itself to the individual.

When those moments are automated, something is lost. The question is whether what is gained, in efficiency, consistency, or accuracy, compensates for what disappears.

What Gets Naturalized

The most significant shifts are often the ones that become invisible. A generation ago, the idea that criminal sentencing might be influenced by algorithmic risk scores would have seemed dystopian. Today, such systems are deployed across multiple jurisdictions, and the debate has moved from whether to use them to how to regulate their use.

This is the pattern that deserves attention. Not the dramatic confrontations between human and machine judgment, but the quiet normalization of algorithmic authority. The moment when a judge stops questioning the recommendation and starts treating it as the default. The moment when the burden of proof shifts from the system to the human who wants to override it.

Springer Nature's research portal hosts numerous studies examining how AI systems interact with legal principles, and a recurring theme is this process of normalization. Systems that were introduced as decision-support tools gradually become decision-making tools. The human in the loop becomes a formality, a signature on a document that was written elsewhere.

The Question of Authorship

Who is responsible when an algorithmic system makes a decision that harms someone? This is not a new question, but the volume under discussion approaches it from an angle that policy debates often miss. The issue is not simply liability, which can be allocated through legal mechanisms. The issue is authorship.

Legal systems assume that decisions have authors: identifiable agents who can be held accountable, who can explain their reasoning, who can be sanctioned for errors. Algorithmic systems complicate this assumption. The developer who trained the model, the organization that deployed it, the official who relied on its output: each can point to the others. Responsibility diffuses.

This diffusion is not accidental. It is a feature of how algorithmic systems are designed and deployed. The complexity that makes them powerful also makes them opaque. The distribution of development across teams and organizations makes it difficult to locate a single point of accountability.

What Comes Next

The rule of law is not a fixed set of requirements but an evolving tradition. It has adapted before: to the rise of administrative agencies, to the expansion of executive power, to the internationalization of legal authority. Perhaps it can adapt again.

But adaptation requires clarity about what is at stake. The risk is not that algorithmic systems will replace legal systems entirely. The risk is that they will transform legal systems in ways that preserve the form while hollowing out the substance. Courts will still exist. Judges will still preside. But the locus of decision-making will have shifted to systems that operate according to logics that the rule of law was not designed to govern.

This volume does not offer solutions. It offers something more valuable: a clear articulation of the problem. For policymakers, technologists, and governance scholars working on AI regulation, that clarity is essential. Before asking how to regulate algorithmic systems, it helps to understand what those systems do to the normative foundations on which regulation depends.

The Human × AI Content Hub continues to track these questions as they unfold across the European AI ecosystem, where the tension between innovation and accountability remains unresolved.