What remains for judges and lawyers once machines verify the law?
Decomposition, not automation: separating what a machine can check from what belongs to judgment.
13 pieces · 4 theses · 1 frameworks · Program RSS

The public debate asks whether machines can replace judges. The Institute thinks that is the wrong question. A judicial decision combines two different kinds of work: the mechanical verification of legal premises (does this authority exist, does it hold what it is cited for, is it still good law, does it govern here) and the discretionary calibration of procedure (how readily to end a case early, at which stage the filter bites, how much fact development to allow, how actively to manage the litigation). Machines are becoming good at the first. The second is much of what judges are for, and it deserves rigorous description rather than either fear or hype.
This program studies what remains for judges and lawyers once verification is cheap. Its measurement side builds on Ariadne's Thread, whose four dispositional axes (termination intensity, kill-point timing, fact-development cost and managerial activism) name the dimensions along which lawful judges differ, and whose durable kill point asks where cases actually die once the uneven survival of procedural rulings on appeal is taken into account. Its constitutional side is Gibsonia: a fictional jurisdiction that enacts its interpretive canons before it allocates any power, so that method comes before power. By making the method of interpretation positive law, the Gibsonia work reframes the Hart-Dworkin debate and asks whether Dworkin's Hercules, the judge who reads the whole tradition and finds the one right answer, becomes computationally tractable once the canons are closed and ordered. Ariadne supplies the other half: where the law does not fix one answer, the honest instrument measures the spread instead of choosing.
Why this matters: if machines verify and humans decide, the legitimacy of the result depends on drawing that line in the open. Decomposition, not automation, is the working principle here: say exactly which parts of a legal decision a machine can check, show the residual that belongs to judgment, and never dress a discretionary choice in the language of legal compulsion.
Theses from this program
Formalized in this program
Reading order
- 01The Long Road to Applied Computational LawSix decades of formalizing legal reasoning, and the question of authority waiting at the end of the road
- 02What Remains for the Judge When the Machine Has Already Verified the Law?
- 03From Hercules to AriadneHow a Half-Century Quarrel in Jurisprudence Became a Quantity, and What the Quantity Cannot Settle
- 04Method Before PowerWhy a Civilization Would Bind Its Interpreters Before It Chose Them
- 05A Morning in the ChamberAn Interpreter, Two Households, and the Canon She Could Not Use