An AI Read 9,000 Supreme Court Dissents To Find Out Which Justices Play Nice On Paper

An AI Model Found the Justices Who Soften Their Dissents With a Compliment First

Supreme Court justices are supposed to be careful, measured writers. A new study reveals a pattern hiding inside their opinions: some justices routinely compliment their opponents’ reasoning even while tearing it apart, and researchers built a machine learning tool that can detect this pattern across decades of legal writing.

Donald Snyder and Caitlyn Pierce, political scientists at the University of Massachusetts Amherst, call it an “appeasement” measure. It tracks moments when a justice writing a dissent or concurrence acknowledges the majority got something right, even while arguing the majority reached the wrong conclusion. Published in American Politics Research, the study finds this on-paper courtesy isn’t random. Certain justices, and certain moments in a case’s history, produce more of it than others.

Consider two responses to Dobbs v. Jackson Women’s Health Organization (2022). Justices Breyer, Sotomayor, and Kagan wrote a blistering joint dissent, calling the majority’s historical analysis “window dressing.” Chief Justice Roberts, writing separately, called the majority’s reasoning “thoughtful and thorough,” agreeing prior precedent should be discarded while arguing the majority went further than necessary. Both disagreed with the outcome. Only one complimented the other side first.

An AI Model Sorted 60,000 Passages Into ‘Concessions’ and Not

Measuring rhetorical courtesy across decades of legal writing isn’t something a researcher could easily do by hand. Snyder and Pierce turned to LEGAL-BERT, an AI model trained on legal text, to scan 9,049 dissents and concurrences written between 1946 and 2011, pulling every passage where a justice directly addressed “the Court” or “the majority.” That search turned up 60,637 passages.

Researchers hand-labeled a sample of those passages as either a “concession” (the justice grants the majority has a point) or a “non-concession” (no ground given), and a second reader checked the work, agreeing with the labels 97% of the time, confirming people could reliably spot the pattern. The AI model, once trained on those examples, got just as good at it, correctly catching real concessions over 90% of the time and landing above 98% overall accuracy on cases it had never seen.

Each justice’s appeasement score, whether for a single opinion or a whole career, comes down to how often they conceded versus how often they didn’t.

‘Wild Bill’ Douglas Scored Lowest, Rehnquist Changed After Becoming Chief

Some findings match reputations Court watchers already knew. Justice William O. Douglas, nicknamed “Wild Bill,” was a prolific opinion writer known for blunt disagreement. His appeasement scores ranked consistently among the lowest in the dataset.

Justice William Rehnquist showed a different trajectory. As an Associate Justice, his scores stayed fairly stable. After his 1986 promotion to Chief Justice, his scores grew more variable but reached new highs, a shift researchers see as possibly tied to his stake in protecting the Court’s legitimacy, though the paper stops short of calling it proven.

At the case level, Justice William Brennan’s 1976 dissent in Francis v. Henderson, the furthest-left justice on the Court that year by the study’s measure, ranked among the lowest appeasement scores recorded. The case involved a petitioner protesting the exclusion of African Americans from the grand jury that indicted him, and Brennan’s dissent offered sharp condemnation with little rhetorical softening.

Justice Harry Blackmun’s concurrence in the 1980 case United States v. Ward, a technical dispute over environmental penalties, scored on the opposite end. Researchers suggest the subject matter may have left room for measured, cooperative language.

Justices at the Ideological Extremes Conceded the Least

A central finding involves ideology. Justices at the ideological extremes conceded less to the majority, while justices closer to the center conceded more, using a standard political science measure that places justices on a spectrum from liberal to conservative. A separate measure that tracks harsh or emotional language showed no such pattern, suggesting something specific: extreme justices aren’t necessarily angrier on the page, they’re just less willing to hand the other side any credit at all.

Coalition size mattered too. As the majority grew larger and more diverse, dissenting justices tended to write more conciliatory language, perhaps because a broader coalition is harder to argue against.

Kavanaugh Says He Writes With the Losing Side in Mind

Snyder and Pierce argue the tool extends beyond legal scholarship. Congressional floor debates and campaign rhetoric involve the same choice: whether to validate any part of an opponent’s argument or reject it outright. During a 2023 Notre Dame Law Review symposium, Justice Brett Kavanaugh described writing with multiple audiences in mind, including the losing party. “I want them to read it and say, ‘Well, okay. They at least treated our arguments with respect and fairly,'” Kavanaugh said.

This appeasement measure offers one way to track how justices frame disagreement in print. It does not show whether conciliatory language improves relationships or strengthens the Court itself. What it does show is that some justices reach for a compliment before delivering a rebuttal, a choice that follows real patterns.

---

Disclaimer: This article summarizes findings from a peer-reviewed study and is intended for general informational purposes. It does not represent legal advice or an official interpretation of any court ruling.

---