Dworkin's Hercules Brought to Life: An AI-Enabled Integrity Engine
For much of modern legal scholarship, serious research has required an enormous amount of intellectual labor devoted not to creating a new argument, but to reconstructing the arguments that came before it. Scholars search for statutes, cases, books, and articles; trace the development of doctrines; summarize competing positions; and document every proposition through elaborate systems of citation.
This work has never been merely clerical. Citation situates an argument within a legal and intellectual tradition. It allows readers to trace authorities, examine competing views, and test whether an author's characterization of the law is fair.
Yet artificial intelligence now forces us to ask a more fundamental question: what exactly is valuable about this activity?
An academic proposition should ultimately stand or fall on the quality of its evidence and reasoning. The fact that another scholar has previously made a similar statement does not make the statement true. Authority cannot substitute for logic.
In the natural sciences, the distinction is relatively easy to understand. A citation may point to an experiment, a measurement, or a body of empirical data that possesses evidentiary value independent of the reputation of its author. The underlying observation can be reproduced, criticized, or incorporated into further inquiry.
Law is different.
A legal proposition rarely derives its meaning from a single piece of evidence. It exists within a network of statutes, precedents, institutional practices, historical developments, interpretive principles, and competing theories. A judicial decision may remain formally valid while its practical significance changes dramatically after subsequent cases or legislative amendments. The same sentence from the same precedent may carry different legal weight at different moments in time.
This makes legal citation both indispensable and deeply imperfect.
A heavily cited legal argument is not necessarily an objective one. Authorities can be selected strategically. Favorable precedents may be emphasized while inconvenient ones are distinguished, minimized, or simply omitted. An older proposition may continue to appear in textbooks and briefs long after the legal environment that originally gave it meaning has changed.
The deeper problem, therefore, is not citation itself.
It is the opacity of the structure that citation attempts to represent.
Hercules and Law as Integrity
This problem brings us naturally to Ronald Dworkin.
Dworkin rejected the idea that difficult legal cases could be resolved simply by locating an existing rule or by exercising unconstrained judicial discretion when no clear rule existed. His theory of law as integrity instead asked judges to interpret legal practice as a coherent whole.
The judge should seek an interpretation that both fits the institutional history of the legal system and justifies that history in terms of principle. Past statutes and judicial decisions are therefore not isolated commands. They form part of an ongoing interpretive practice in which present decisions must be understood in relation to what came before them.
To illustrate this ideal, Dworkin imagined a fictional judge whom he called Hercules.
Hercules possesses extraordinary patience, knowledge, and intellectual ability. Faced with a difficult case, he can examine the entire relevant body of legal materials and construct the interpretation that best reconciles them according to principle. He does not merely count precedents or follow whichever authority appears most convenient. He attempts to understand the law as an integrated structure.
Hercules was, of course, an idealization.
No human judge can read every relevant case, reconstruct every doctrinal development, consider every plausible factual distinction, and continuously test every proposed interpretation against the entire history of the legal system. The cognitive burden is simply too great.
But that assumption may no longer be as obvious as it once was.
Artificial intelligence does not make Dworkin's Hercules literally real. Nor should an AI system be treated as a philosopher-judge whose answers deserve authority merely because they are computationally sophisticated.
What AI may make possible, however, is something more interesting: a technological infrastructure capable of performing part of the informational and analytical work that Hercules was imagined to perform.
It may allow us to build an integrity engine.
From Footnotes to Argument Structures
Traditional legal writing compresses a complicated structure of authority into a surprisingly primitive interface: the footnote.
A proposition appears in the text, followed by one or several citations. Those citations tell the reader where some supporting material can be found, but they rarely reveal the full structure surrounding the proposition.
Has the cited case subsequently been limited?
Has the statute on which it relied been amended?
Do other courts disagree?
Is the proposition dependent on a narrow factual distinction?
Has later scholarship identified an inconsistency in the reasoning?
Does the authority remain technically valid while becoming practically obsolete?
A diligent researcher can investigate all these questions, but doing so requires reconstructing a network that is largely invisible in the finished text.
AI creates the possibility of reversing this relationship.
Instead of forcing the reader to reconstruct the network behind a citation, a legal information system could make that network directly visible.
A legal proposition might be accompanied by a structured representation of:
- the principal authorities supporting it;
- the current legal status and relative authority of those sources;
- statutory amendments affecting the analysis;
- subsequent decisions that extended, limited, distinguished, or questioned earlier precedents;
- competing lines of authority;
- factual distinctions that may change the legal outcome;
- significant scholarly disagreements; and
- the historical development of the doctrine.
Citation would no longer function merely as a pointer to documents.
It would become an entry point into an argument structure.
The question would shift from:
"What authorities support this proposition?"
to:
"Where does this proposition stand within the entire legal landscape?"
That is a much more demanding question. It is also much closer to Dworkin's conception of integrity.
Making Legal Confidence Inspectable
The goal of such a system should not be to produce more confident answers.
It should be to make confidence itself inspectable.
Suppose an AI system concludes that a particular interpretation is currently the strongest reading of Korean law. A conventional AI assistant might provide the conclusion, summarize several cases, and attach citations.
An integrity engine should do more.
It should reveal why the conclusion appears stronger than its alternatives. It should show which authorities support it, which authorities resist it, whether the relevant doctrine has shifted over time, which factual differences are decisive, and where genuine uncertainty remains.
The legitimacy of the system would therefore not arise from the authority of artificial intelligence.
It would arise from transparency.
This distinction is essential.
An AI-generated legal answer can easily reproduce the weaknesses of ordinary advocacy. It can select supportive materials, construct a persuasive narrative, and present the result with impressive fluency. Indeed, generative AI may make selective legal reasoning easier rather than harder if its principal objective is simply to produce convincing answers.
An integrity engine would be designed around a different objective.
Its task would not be merely to answer.
Its task would be to expose the structure within which an answer can responsibly be made.
From Legal Database to Integrity Engine
This perspective also suggests a different future for established legal information providers.
A platform such as LawnB, for example, may possess value that is difficult to recognize if its content is viewed simply as a collection of documents. Its accumulated statutes, judicial decisions, commentaries, academic writings, and practical materials represent decades of legal development.
The deeper asset may not be the documents themselves.
It may be the relationships hidden among them.
If those relationships can be reconstructed, validated, and dynamically presented, a traditional legal information service could evolve through several conceptual stages:
Document Repository → Legal Knowledge Graph → Argument Graph → Integrity Engine
A document repository allows users to retrieve sources.
A legal knowledge graph connects those sources through concepts, courts, statutes, citations, legal issues, institutional relationships, and time.
An argument graph goes further by representing how particular propositions are supported, challenged, qualified, distinguished, or superseded.
An integrity engine adds another layer. It asks how a proposed legal interpretation fits within the broader structure of the legal system and makes that assessment open to inspection.
At that point, legal research ceases to be primarily an exercise in finding documents.
It becomes an exploration of the architecture of legal reasoning.
The Changing Value of Legal Scholarship
Such a transformation would also change what we mean by scholarly labor.
The mechanical cost of finding and summarizing prior literature will almost certainly decline. AI systems are already becoming capable of retrieving large bodies of material, identifying recurring propositions, and generating preliminary syntheses at speeds that no individual researcher can match.
This does not mean that scholarship becomes less important.
It means that its center of gravity may shift.
The difficult intellectual task will increasingly be to determine which relationships actually matter: whether two cases genuinely stand for the same proposition, whether a later judgment modifies an earlier doctrine or merely applies it to different facts, whether an apparent conflict reflects a deeper conceptual distinction, and whether a contemporary interpretation remains coherent with the broader development of the law.
Citation therefore need not disappear in the age of AI.
It may become more demanding.
The purpose of citation may gradually shift from demonstrating that an author has consulted the relevant literature to making the architecture of a claim visible.
A good scholarly work would then be valuable not simply because it contains many authorities, but because it reveals how those authorities relate to one another and why those relationships matter.
Integrity as an Antidote to Selective Law
There is also a broader institutional consequence.
Law is unusually vulnerable to selective presentation.
Lawyers are advocates. Governments defend policies. Scholars develop theories. Litigants seek favorable interpretations. Each has perfectly understandable reasons to emphasize some authorities over others.
The problem arises when selective presentation begins to masquerade as a complete description of the law.
A sufficiently transparent legal information system could make this more difficult.
Imagine that whenever someone relies on a precedent, the surrounding history becomes immediately visible: contrary cases, subsequent statutory amendments, factual limitations, academic criticism, and later decisions narrowing the original proposition.
Such a system would not prevent disagreement.
Nor should it.
Legal disagreement often reflects genuine conflicts among principles, institutions, and interpretations. An integrity engine should not conceal those disagreements beneath an artificially synthesized consensus.
Its contribution would instead be to make disagreement more honest.
It would increase the cost of ignoring inconvenient legal materials because those materials would be structurally connected to the proposition being asserted.
In this sense, AI could contribute to the rule of law not by deciding cases for us, but by making the legal environment in which decisions are made radically more transparent.
Hercules Brought to Life
Dworkin's Hercules was never meant to be a technological prediction. He was a philosophical device: an ideal judge capable of taking the legal system seriously as an integrated history of principles and decisions.
Yet AI gives the metaphor a new significance.
We may never build a machine that possesses the judgment of Hercules. And perhaps we should not try.
What we can plausibly build is something that gives human lawyers, judges, scholars, and citizens access to capabilities that were previously available only to Dworkin's imaginary judge: the ability to examine vast bodies of legal material simultaneously, trace their relationships across time, identify inconsistencies, surface opposing authorities, and test interpretations against a much broader institutional history.
In that sense, Hercules need not become an AI judge.
He can become an infrastructure.
The most valuable legal AI of the future may therefore not be a system that tells us what the law is with ever greater confidence.
It may be a system that allows us to see why we believe the law to be what it is, what evidence supports that belief, what challenges it, and where judgment must still begin.
That is the promise of an AI-enabled integrity engine.
Not the replacement of legal judgment, but the creation of an environment in which legal judgment can be exercised with a degree of transparency, completeness, and intellectual integrity that was previously beyond human reach.