Against whatever this is


The glowing red eye of HAL-9000
Credit: Pixabay (CC0)

I wrote a post the other day called “Against Generative AI in Legal Writing.” These days I find it very easy to say what I am against. I am not going to give a list here. All right, I am going to give a partial list in a footnote.1I am against incompetence. I am against cruelty. I am against cowardice and lack of self-respect. I am against poseurs. I am against ideologues and also against people with no ideas except self-interest. I am against lack of respect for history and tradition and even more against ignorance of history and tradition. I am against bullies. It is harder to say what I am for. I would like to give you a rousing post about what I am for, and perhaps I will as we get a bit of distance from the most recent US elections. But when I saw an absurd paper on SSRN, We Built Judge.ai. And You Should Buy It, I knew that I’d found one more thing that I’m against and that like Casey at the Bat, I couldn’t let a paper like this go by without taking a swing at it, even knowing that Casey’s strikeout is a warning to anyone who expresses deep skepticism at legal AI in 2025.

I should start by saying that I am not 100% sure that the paper is not an elaborate exercise in trolling either AI hypemongers or some legal scholars—I’ve never met anyone like this but the authors seem to think they exist—who think that the process of deciding even private law disputes can be reduced to an algorithm. If the paper is an exercise in trolling, then I congratulate the authors: well done!

But I’ll assume that the authors are in earnest. Who are the authors, by the way? A 3L at Harvard Law School, a recent graduate who according to LinkedIn is now working as a federal law clerk, and someone a few years out of college who earned a master’s degree in computer science a couple of years ago. In other words, people who have, apparently, never participated in a legal dispute as counsel or as decisionmakers, and maybe never as litigants. I do not think it is an ad hominem attack to say that these are not the people who have the experience that would suit them to design a new system of dispute resolution.

They have invented, and applied for a patent for, a method of automating arbitration that relies on an AI “decisionmaker” they call arbitrus.ai. You can read the (semi-) technical description in the paper. Here is the heart of it, which looks to me like like it was taken from the patent application:

Users are formally invited to the Arbitrus.ai portal only after a Party files a claim (“Party A”). (300). Arbitrus.ai then has seven days (7) to serve valid, legal notifications to each party (301); valid legal notification includes the substantive complaint from Party A. Each party then has seven (7) days prior to confirm notification (302-303). Once notification is confirmed, the parties are advanced to the second stage: briefing (343). During Briefing, the deadline for Party B’s response brief is seven (7) days (305-308). Party A then has seven (7) days to file an answer (308-309). During this period, either Party may file a dispositive motion (306-307, 313, 317). Arbitrus.ai then decides the motion (310-312, 316,318-319, 321); if a motion is dispositive, a ruling is issued and final order authorized (338-339, 342). If a motion is not dispositive, a ruling is still issued, and the Parties sent back for Briefing (312, 321). After Briefing, Parties are sent to Discovery (344).

During Discovery, both Parties are ordered to submit evidence (315, 323, 324-325). After Discovery is processed, Parties have seven (7) days to file objections and request additional evidence (328); if Arbitrus.ai agrees with a Party’s objection or request for additional information, the Parties have seven (7) days to resubmit relevant Discovery to be processed (322). Discovery is then processed again. If there are objections, and Arbitrus.ai agrees, then the process begins again (322). If there are no objections, Arbitrus.ai may decide to order additional briefings, in which case the process begins again (330-331, 341-342). Alternatively, if Arbitrus.ai determines no additional briefing is necessitated, it orders a Hearing (333); during which time, both Parties are simultaneously queried for questions by Arbitrus.ai (334-336). Both Parties will have an opportunity to review responses. After the Hearing is concluded, Arbitrus.ai will issue a ruling within three (3) days (337-339). Either as part of, or separate to, the Ruling (depending on the case), Arbitrus.ai will also issue a Final Order (342), that contains the specific award.2If you are wondering about the numbers in parentheses, you have not read too many patents. They are references to elements of a diagram that is reprinted in the paper and was probably taken from the patent application showing the steps of the method the authors seek to patent. Exception: the authors have the bad habit, which many lawyers share, of putting numerals in parenthesis after numbers in legal documents: instead of “the Parties have seven days,” or “the Parties have 7 days,” they choose to write, “the Parties have seven (7) days.” Why?

There are a lot of ways to attack this misguided venture. Here are the ones that seem most salient to me.

  • The paper explains that in the “discovery” phase of the process, the parties will take depositions of the witnesses, which will include cross-examination. Great, but how will the AI decisionmaker decide what testimony to credit? “In the future,” the authors write, “separate IP will need to be created to referee cross-examination.” A system of dispute resolution that has no way to listen to people’s stories and then decide what happened strikes me as not really a system of dispute resolution at all. And in principle, I do not see how anyone other than a human being can listen to conflicting testimony and decide what is true and what is false. Even when testimony does not squarely conflict, often the law’s standards are not bright lines. You can see this in some of the test cases cited in the paper that the AI decisionmaker “decided.” In one case, about an employee’s right to unpaid severance after termination, the employer claimed the employee was guilty of gross mismanagement had breached her fiduciary duty by, for example, creating a “toxic” workplace environment and had exceeded budget without authorization. Even if there were a bright line separating toxic and healthy workplace environments or separating “gross” mismanagement form ordinary mismanagement, and even if there is a case where an appellate court had affirmed a judgment finding that conduct like the employee’s conduct supported a finding of termination for cause, that doesn’t mean that a fact-finder is required to reach the same conclusion in another case. There is not a single right answer to such cases under our law, any more than there is a single right answer to whether particular conduct is negligent. Otherwise, in court, every case would be decided on summary judgment. You could go down the list of sample cases discussed in the paper and make the same point again and again. The authors seem to acknowledge this when they acknowledge that the only people who will be obviously on board with this project are “formalists,” people who think the law should be utterly predictable in its outcomes. I don’t think I know anyone who has experience with real, messy, human disputes who thinks that way or who thinks it is reasonable to think that way.
  • If two sophisticated businesses want to agree to use a system like this, I have no objection in principle, any more that I would object to an agreement to resolve a dispute by flipping a coin. I doubt that real-world businesses will want to have their disputes decided in this way. But the real issue is not with parties who go in with their eyes open. The real issue is that arbitration requires the parties’ consent, but that in many contract the consent is merely notional. The paper recognizes that consumer arbitration poses a problem, because consumer agreements to arbitrate are typically contracts of adhesion. Here is their answer: “Well … there really isn’t one. Arbitration isn’t sunshine and rainbows; it’s largely designed so that you lose. With Arbitrus.ai, you just lose faster and more transparently.” First, of course, they are wrong to say that arbitration is designed so that consumers lose. The real problem for consumers is not the forum where disputes are decided; it is the substantive terms of the contracts, which often limit or exclude remedies. It is unclear to me why anyone thinks that a court is more likely to be biased toward a consumer (or to put it another way, less likely to be biased towards a business) than an arbitrator, and if the thought is that a jury might be more inclined to favor the consumer, claims that are legally untenable generally do not make it to a jury. If I may provide an anecdote, I’ll tell you that I am appointed as arbitrator in both consumer disputes and commercial disputes. If you ask me how many of the consumer cases I’ve decided came out in favor of the consumer and how many in favor of the business, I would have no idea. If you ask me whether I worry that a decision in favor of a consumer could lead to a loss of business, I would laugh and tell you that consumer cases are not really money-makers anyway, as they are done at reduced rates; I treat that work as a kind of public service and could care less whether a particular business or a particular lawyer decides not to appoint me in a second case after losing the first. But I will tell you that I was appointed in a case where the business was represented by a lawyer who had appeared before me in a prior consumer case—an unusual case to be sure—where I had awarded the consumer very significant damages. If you ask why arbitrators are likely to enforce contract provisions that seem to favor businesses over consumers, the answer is that there is a lot of judicial precedent that does likewise.3Of course I understand there is a structural concern about “repeat players” in arbitration, which suggests an incentive for arbitrators to favor the business over the consumer. But if you think that courts do not also compete for business from repeat players, I suggest you google “Common Pleas” and “King’s Bench.” To be fair, I do think we are in an unusual time when American courts sometimes seem to want to drive away business, while courts elsewhere are seeking to attract business through, for example, the creation of commercial courts. But that is a topic for another day.

I am not going to bother with the human and moral objections to this kind of venture, as it is clear from the paper’s text that the authors don’t care about such things. “We’ve never claimed that any of this is rooted in morals. Rather, it’s rooted in the modern enlightenments of formalism and free markets where efficiency is God and the machine is its vicar on Earth.” Fortunately people who think like this and talk like this don’t run the … well, I suppose things that I thought were true may not be so true in 2025. But we can still resist the bogus tech and libertarian triumphalism that this paper exemplifies.4There are various less central problems with the paper that I will not catalogue. One that I find particularly annoying because it reflects a widespread misunderstanding: the American Arbitration Association is not an “arbitration firm” that “employs stables of arbitrators.” AAA arbitrators are self-employed, or employed by their law firms, and the AAA is a non-profit body that provides administrative services in arbitrations, including in the appointment of arbitrators.

  • 1
    I am against incompetence. I am against cruelty. I am against cowardice and lack of self-respect. I am against poseurs. I am against ideologues and also against people with no ideas except self-interest. I am against lack of respect for history and tradition and even more against ignorance of history and tradition. I am against bullies.
  • 2
    If you are wondering about the numbers in parentheses, you have not read too many patents. They are references to elements of a diagram that is reprinted in the paper and was probably taken from the patent application showing the steps of the method the authors seek to patent. Exception: the authors have the bad habit, which many lawyers share, of putting numerals in parenthesis after numbers in legal documents: instead of “the Parties have seven days,” or “the Parties have 7 days,” they choose to write, “the Parties have seven (7) days.” Why?
  • 3
    Of course I understand there is a structural concern about “repeat players” in arbitration, which suggests an incentive for arbitrators to favor the business over the consumer. But if you think that courts do not also compete for business from repeat players, I suggest you google “Common Pleas” and “King’s Bench.” To be fair, I do think we are in an unusual time when American courts sometimes seem to want to drive away business, while courts elsewhere are seeking to attract business through, for example, the creation of commercial courts. But that is a topic for another day.
  • 4
    There are various less central problems with the paper that I will not catalogue. One that I find particularly annoying because it reflects a widespread misunderstanding: the American Arbitration Association is not an “arbitration firm” that “employs stables of arbitrators.” AAA arbitrators are self-employed, or employed by their law firms, and the AAA is a non-profit body that provides administrative services in arbitrations, including in the appointment of arbitrators.

Leave a Reply

Your email address will not be published. Required fields are marked *

Thank you for commenting! By submitting a comment, you agree that we can retain your name, your email address, your IP address, and the text of your comment, in order to publish your name and comment on Letters Blogatory, to allow our antispam software to operate, and to ensure compliance with our rules against impersonating other commenters.