Beyond chatbots: Is there a role for adjudicative AI?

By any measure, California's trial courts are busy. Based on an informal survey supported by a few administrative record requests that I recently conducted of courts of similar size to my own, caseloads in civil direct calendar departments may approach or exceed 1,000 per judge. At the time of writing, my own civil caseload numbers approximately 890 unlimited cases and 40 limited cases, which I handle with the support of one-third of a legal research attorney (shared with two other assignments). By contrast, a federal district judge may carry a docket of roughly 350 cases, typically supported by multiple law clerks.

This reality matters because a transformation is already underway.

For the past two years, the legal profession has understandably focused on generative artificial intelligence as a tool used by lawyers. We have debated hallucinated citations, confidentiality concerns, disclosure obligations and the ethical responsibilities accompanying AI-assisted legal practice. While those conversations remain important, they are not the only conversations we should be having.

Artificial intelligence is rapidly reducing the time and cost required to produce legal work product. Research and drafting that once required hours or days may now be completed in minutes. Smaller firms now have access to capabilities once available only to the largest firms. Self-represented litigants are gaining access to increasingly sophisticated legal tools.

Whether welcomed or resisted, these developments are likely to change the economics of litigation. As the cost of producing legal work product declines, it is reasonable to expect that more disputes will be litigated; more motions will be filed and increasingly sophisticated advocacy will become available to a broader range of litigants.

Every efficiency gained on one side of the courthouse door creates additional demands on the other. If that prediction proves correct, we need to be thinking now about how the courts should respond. The stakes are high. A failure to respond adequately could further impair access to the courts and, ultimately, erode public confidence in their legitimacy.

We appear to be at an inflection point. Two levers are obvious. The first is the traditional one: increase the number of judges, legal research attorneys and court support staff. California's courts need additional judicial resources today, regardless of advances in artificial intelligence.

The second is more difficult to discuss: whether courts should use what I will call "adjudicative AI," and if so, to what degree. I do not mean the AI-assisted research tools already used throughout the legal profession. I mean tools that participate more directly in the adjudicative process, under carefully defined conditions, with appropriate transparency, meaningful human oversight and ultimate judicial responsibility for every final decision.

I am not suggesting that the courts should embrace the full range of possible adjudicative uses of AI--or necessarily embrace adjudicative AI at all. Instead, my point is that courts must have that conversation even as deployment begins. And doing so requires confronting the source of much of today's anxiety around AI's role not just in the courtroom, but on, or at least adjacent to, the bench.

That anxiety exists and is powerful, as reflected in the reporting around pilot projects underway in some California courts. (See, e.g., Cayla Mihalovich & Khari Johnson, "California Judges Are Testing a New AI Clerk, and You Won't Know if It's Looking at Your Case," LAist (May 26, 2026); Mark S. Adams and Sharon R. Klein, "AI's impact in California courts and why judges and the legislature are at odds," Daily Journal (Apr. 6, 2026); James Queally, "AI Pilot Program in L.A. County Courts Will Help Judges Craft Rulings in Some Cases," Los Angeles Times (Mar. 18, 2026).)

The concerns are expressed both within and without the courthouse. We see the ongoing problems with hallucinations, including fake authorities and fabricated quotations. Add to that reality the well-founded concerns about biases baked into AI's knowledge base that may surface in nonobvious ways. And if not deployed in an ethical and responsible way, AI may challenge a fundamental premise of our system and the ethical strictures that support it: that judges make decisions solely based on the evidence in the record and the arguments of counsel before them. (See, e.g., the California Code of Judicial Ethics, Canons 2A [judge must "act at all times in a manner that promotes public confidence in the integrity and impartiality of the judiciary"] & 3B(7) ["[A] judge shall not independently investigate facts in a proceeding and shall consider only the evidence presented or facts that may be properly judicially noticed."]; cf. Evid. Code §§ 455, 459(d).)

The concerns are legitimate. Secrecy magnifies them. Anxiety may be at its highest, and public resistance may be at its greatest, when the public is completely in the dark as to how AI might be deployed. The spectrum of potential uses runs from preparing summaries and organizing data to the far more consequential use of generating proposed analyses, rulings or orders. The latter raise the most provocative questions and challenges to our current conceptions and the public's expectations about judging. In all cases, the public is right to wonder: Will judges independently evaluate the briefs, the evidentiary record and AI's work? What safeguards will exist? The early reporting does not reflect these issues being publicly addressed.

When technology enters the judicial process as a black box, uncertainty naturally gives rise to suspicion. That should not surprise us.

Public confidence in the judiciary depends upon legitimacy. It requires more than a correct result; it requires sufficient understanding of the process to trust that judicial responsibility has not been delegated to an opaque system. When courts cannot explain the role technology plays in the court's business, the public is left to imagine the possibilities for itself--and those imagined possibilities may be more troubling than the reality.

Given the resource challenges, courts may increasingly confront not only whether AI should be integrated into the adjudicative process, but under what conditions. Several principles seem essential.

First, transparency. To be sure, the process of judging has not been fully transparent historically; by way of example, how judges work with judicial law clerks and research attorneys varies and is generally not disclosed. Nonetheless, at the institutional level, the capabilities of legal AI products used by the courts should be public, along with the technological and ethical guardrails governing their use. In individual cases, judges should consider disclosure, particularly when that use exceeds what the litigants might reasonably expect. (Cf. Cal. Stds. Jud. Admin., std. 10.80.)

Second, human responsibility. Artificial intelligence should assist judges, not replace them. Every final judicial decision must remain the responsibility of a human judge.

Third, meaningful human oversight. AI-generated analyses should never become rubber-stamped recommendations. Judicial review must remain genuine rather than perfunctory.

Fourth, careful deployment. Not all AI is suited to judicial work. Courts should deploy only tools validated for the task at hand--evaluated for accuracy and reliability in the specific adjudicative context--rather than relying on unconstrained general-purpose systems merely adapted for judicial use. Any such tool should be constrained to the record and governing authorities properly before the court and not surface extra-record information drawn from its own training data.

Fifth, empirical evaluation. Accuracy, consistency, litigant confidence, appellate outcomes, potential disparate impacts, transparency, efficiency and judicial workload should all be evaluated on an ongoing basis.

Sixth, judicial AI literacy. Courts cannot responsibly govern technology that they do not understand. Ethical and responsible use of powerful AI tools requires judges to understand, at least at a functional level, how AI works, where it performs well, where it is prone to error and where its limitations require careful human judgment.

These principles are not intended to answer every question. They are offered as a means to help ensure that the right questions are asked as use develops and expands.

Artificial intelligence will continue to advance in the legal space. Lawyers will continue using it. Litigants will continue relying upon it. Courts will increasingly confront it.

The question is no longer whether these developments will occur. The question is whether we will shape them through thoughtful governance or instead react to them after the fact.

Author's Note: In the interest of transparency, I used legal AI tools during the research and development of this article. I independently reviewed the authorities. I accept full responsibility for every statement and conclusion expressed here.