Return to the cave: AI and the protection of absent parties

Thirty-two years ago, I found myself in Plato's cave with Judge Jack B. Weinstein. The cave was, of course, metaphorical--a reference to Plato's allegory and the title of a law review article we wrote examining the challenges judges face when reviewing proposed class-action settlements. (See Jack B. Weinstein & Karin S. Schwartz, "Notes from the Cave: Some Problems of Judges in Dealing with Class Action Settlements," 163 F.R.D. 369 (1995).)

In a series of mass tort cases, spanning asbestos to Agent Orange to DES, Judge Weinstein had developed the class action as a tool for mass justice. But he was concerned about a practical problem: that posed by effectuating substantial justice, and particularly in protecting the interests of absent class members, when a class action settles. Once the parties reach agreement, the incentives in the courtroom change. Class counsel, the class representatives and the defendant all have reasons to seek approval of the settlement. The absent class members--whose rights also will be affected--have no one independently advocating for them. A judge must evaluate the negotiated agreement without having witnessed the negotiations that produced it. Much of what matters occurred outside the courtroom. Like prisoners in Plato's cave, judges see only the shadows cast by events they did not observe directly. And so that early article identified potential tools for judicial review of class-action settlements, such as increased judicial involvement in the settlement process, appointment of guardians ad litem for absent class members, and conditioning court approval on changes to the proposed settlement.

Fast forward to 2026. The landscape has changed dramatically. As a superior court judge assigned to a direct calendar civil department, I may see as many as two motions for class-action settlement approval a week, as compared to the one-to-two class-action settlements per year that I recall from my years as a law clerk in the 1990s.

The fundamental tension has not changed. What has changed is the volume. How can a judge managing nearly a thousand cases apply the searching review Judge Weinstein envisioned to every uncontested matter affecting absent parties? Yet those absent parties' rights are no less important simply because the docket has grown.

This is where AI may have an important role to play: supporting judges in reviewing uncontested matters that affect the rights of people who are not before the court, or who cannot effectively protect their own interests.

Class actions are only one example. Courts routinely confront proceedings in which meaningful adversarial testing is limited or absent. Consider, for example, civil proceedings involving petitions for approval of a minor's compromise, review of default judgment packages, uncontested good-faith settlement determinations and structured-settlement transfer approvals (see Ins. Code § 10139.5).

AI should never decide whether relief should be granted. However, it could perform two distinct and supportive functions. First, it could identify objective defects: missing evidence, inconsistent calculations, procedural omissions or required findings unsupported by the record. Second, it could identify discretionary issues deserving closer judicial attention, presenting both the evidence supporting approval and the evidence suggesting caution. In both cases, the decision remains entirely with the judge. To maximize fairness and transparency, we would benefit from the development of purpose-built AI to support these functions, potentially with disclosure when they are utilized.

Recognizing that not all judges, lawyers and readers of the Daily Journal practice in the civil arena, and recognizing the limits of my own experience, I asked a legal AI system to identify other (non-civil) proceedings in which judges review matters without meaningful adversarial testing. The suggestions below are offered as possibilities for discussion, and not as endorsements or conclusions.

In criminal proceedings, judges must act where adversarial testing is thin or absent: reviewing ex parte applications for search and arrest warrants, where the affected person is not present; taking felony pleas, where the court must independently confirm that the plea is knowing and voluntary and that the required advisements were given, and deciding the many post-conviction and resentencing petitions that arrive unopposed or from self-represented petitioners, where eligibility turns on defined statutory criteria.

On an indigent defendant's first appeal as of right from a criminal conviction, when appointed counsel identifies no arguable issue, the appellate court must itself examine the entire record for error on behalf of a defendant who has no advocate making the argument. (See People v. Wende (1979) 25 Cal.3d 436.)

In juvenile and dependency proceedings, judges make statutorily required findings--including notice and inquiry obligations designed to protect children--whose omission is a recurring source of reversible error.

In probate and mental-health matters, conservatorships (including LPS conservatorships), guardianships, and estate and conservatorship accountings, the judge must protect people who cannot protect themselves.

And in federal bankruptcy, judges perform an analogous role in reviewing reaffirmation agreements, proposed settlements, and fee applications on behalf of debtors and diffuse creditors who are not meaningfully represented in the courtroom.

In these settings, the same structural feature recurs: no adverse party is positioned to surface error, so the responsibility falls to the judge. And to the extent that much of the review is structured and record-based, these contexts may present precisely the conditions in which carefully designed AI might help a judge see what the shadows would otherwise obscure.

Judge Weinstein devoted much of his career to asking how courts could better protect people who were not present to protect themselves. Thirty years later, that question remains. Artificial intelligence does not answer it. But if thoughtfully designed and carefully constrained, it may become another tool for judges seeking to do what he always believed the law should do: look beyond the parties before the court and safeguard the rights of those whose voices are hardest to hear.

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.