In 2025, Eric Posner and Shivam Saran from the University of Chicago’s Law School sought to explore whether artificial intelligence can replace human judges. Reference Posner and Saran ¹ They conducted a series of experiments comparing how judges, law students, and large language models (LLMS) would rule when asked to uphold or overturn the convictions of individuals accused of aiding and abetting war crimes in an international criminal tribunal. The researchers drew upon 2016 and 2024 experiments by Holger Spamann and Lars Klöhn that examined the formal adherence by judges and later students to precedent in comparable cases.Reference Spamann and Klöhn ² The Spamann and Klöhn experiments found that judges were surprisingly less formalistic and committed to precedent than academic lawyers had expected and students had turned out to be.Reference Spamann and Klöhn ³ The Posner-Saran team invited LLMs into the experiment, comparing the judicial and student outcomes against the artificial intelligence models. They found that the LLMs and students were similarly more formalistic in their treatment of precedent than the judges were. They observed that the LLMs adhered strictly to legal rules and “disregard[ed] non-legal factors.” ⁴

Posner and Saran also found that judges were more willing to disregard precedent than the LLMs and the students were particularly where defendants exhibited qualities that could be characterized as “sympathetic.” ⁵ The researchers also noted that even though judges are more likely than either LLMs or students not to follow precedent, they were still likely to characterize their conclusions as being in line with precedent. ⁶ In fact, the researchers asserted: “Although sympathy seemingly influences human judges’ decisions, the judges avoid mentioning it, creating an outward appearance of formalism.” ⁷

Posner and Saran ultimately concluded that their experiment disrupts a perception that is held by judges themselves as well as the larger public, a perception that may also be built into the LLMs: that “the law consists of well-defined rules … judges have little discretion in applying.” ⁸ They noted that the tendency of judges to employ discretion is a potential strength, emphasizing “human judges are able to depart from rules when following them would produce bad outcomes from a moral, social or policy standpoint.” ⁹ In short, these experiments and the researchers’ findings are compelling not just because LLMs are not ready to replace human judges, but also because judges rely substantially on “humanistic” factors to render decisions. And in fact, judicial discretion can be a good thing in that it can be responsive to conditions that the law, itself inherently flawed and replete with gaps, cannot always predict — like pandemics and their health effects. The difficulty with such discretion is that it will inevitably lead to disparate outcomes that may feel incoherent.

This idea animates the tension at the heart of Mooney, Larkin, and Howard-Williams (Mooney et al.)’s paper on compassionate release: reliance on some amount of judicial discretion in the face of statutory and regulatory ambiguities can risk inconsistent and potentially health-harming outcomes for claimants.Reference Mooney, Larkin and Howard-Williams ¹⁰ They write a timely and compelling analysis of federal district court opinions adjudicating compassionate release petitions by incarcerated individuals during COVID-19 pursuant to the First Step Act. These petitions typically turn on the First Step Act’s requirement that circumstances justifying early release are “extraordinary and compelling,”Reference Mooney, Larkin and Howard-Williams ¹⁰ a standard that is inevitably subjective. Discretion is necessary as an antidote to rigid formalism, particularly where compassionate release legislation could not have anticipated the onslaught of COVID-19. But discretion is also imperfect and can give rise to inconsistencies that may be contingent on time or place (jurisdiction). Mooney et al. identify several issues that are important for us to grapple with as the role of district judges in shaping people’s lives is increasingly in sharp relief.

Mooney et al. explore the distinct outcomes for petitioners Alberto Pena and Chadwick Thompson after each separately sought compassionate release while asserting comparable COVID-19 health concerns in the Southern District of New York and the Eastern District of Texas (upheld by the Fifth Circuit). The results in these cases depended on district courts’ perceptions of the First Step Act’s extraordinariness mandate as well as on the facts surrounding each defendant’s health, crime, and conviction. In both instances, the men suffered from but managed hypertension and high cholesterol. In Pena’s case, the court took into serious consideration that he, sixty years old at the time of his petition, had been given a seven year sentence for contributing to but not physically participating in a robbery, had had a limited criminal history with no prior violent offenses, appeared to be well-regarded in his community, was perceived as unlikely to recidivate, and was close to the end of his sentence. ¹¹ On the other hand, in Thompson’s case, the court emphasized that the petitioner, a forty-three year old, had been accused of drug trafficking conspiracy, had pled guilty to conspiracy to possess and distribute drugs, had been deemed a “career offender,” and had served only eight years of a twenty-year sentence. ¹² The outcomes in these two cases involving individuals with comparable health conditions appear reasonable: an older inmate with a modest criminal background, a community that values him, and a short amount of remaining time on his sentence is released, while a younger inmate with a more extensive criminal background and a much longer remaining sentence is not.

But in both cases — and other compassionate release cases Mooney et al. highlight — district judges, typically experts in neither primary care nor public health, face the difficult task of anticipating the potential health outcomes of each individual petitioner. And they must make these decisions without being able to factor in that people react to disease and to potential interventions to mitigate disease, like vaccination, in ways that even healthcare providers cannot predict. Moreover, these judges issue opinions in such matters, we must assume, with some awareness that to grant an early release is to grant a privilege.

Indeed, the final paragraph in the Thompson opinion continues to stay with me, as I reflect on these complex issues:

One might argue that of course Thompson could point to no such case because the pandemic was still new as was the body of pandemic-related compassionate release cases. (Indeed, the court cites only four opinions, two of which are within its circuit, rejecting compassionate release.) The court’s assertion here reads as absolute in a revealing way, as if to suggest that it cannot provide Thompson with a special entitlement based on his fear of COVID. But the idea of “privilege” or “desert” or “entitlement” seems out of place when we are considering people’s health — even the health of incarcerated people who had pled to doing bad things. Moreover, isn’t fear associated with an unprecedented global pandemic that disproportionately infected and killed people in carceral and other congregate settings legitimate? ¹⁴

Thanks to studies like Posner and Saran’s, among other research, we know that district and other judges do not magically transport themselves into a realm of neutral objectivity when adjudicating cases like these; instead, they remain humans, who confront perceptions, values, social and political constraints, and other non-neutral factors when analyzing facts and law.Reference Ryckman ¹⁵ And as Posner and Saran remind us, that can be a good thing. We also know that district judges make decisions under complex, difficult, and, in the face of exigencies like COVID-19, unforeseen circumstances. The question that arises here is whether we should expect them to be more transparent with us and with themselves in general about the ways they are informed by subjective realities. Maybe we are generally better off when judges acknowledge that in fulfilling their obligation to interpret the law, they are also simply doing their best: the law is frequently untested, particularly in the face of a complex and changing world; it is rarely absolute; and decisionmakers need to be flexible. But we have to recognize the many slippery slopes of this transparency: judicial discretion affects individuals (and the communities they come from) in critical and material ways. ¹⁶ But transparency can also pose risks to district judges as they are increasingly subject to political attack. ¹⁷ With these considerations in mind, maybe it means that lawyers, public health practitioners, health workers, court systems, policymakers, and even ordinary people, may be tasked with identifying new ways to provide these judges (and litigants) with better support, more robust information about how the law will directly affect ordinary people’s lives, and potential alternatives when the options feel zero sum. Maybe those efforts, in due time, can also trickle upward to appellate judges.