How will artificial intelligence (AI) change the judiciary, a pillar of American democracy? One popular proposal is to replace human judges with AI systems. It is not difficult to see this proposal’s appeal. AI judges could make adjudication faster, cheaper, and less biased. Some legal scholars take this possibility seriously. Eugene Volokh (UCLA Law) argues that if an AI judge “can create persuasive opinions, capable of regularly winning opinion-writing competitions against human judges—and if it can be adequately protected against hacking and similar attacks—we should in principle accept it as a judge, even if the opinions do not stem from human judgment.”^[1] This isn’t just an academic debate, either. A handful of countries are moving in this direction. A recent Stimson Center report on AI in the judiciary finds that the UAE and Singapore use AI to make legal recommendations, and Egypt, China, Argentina, Brazil, and Colombia have even incorporated “AI semi-decision-making systems.”^[2]

Most legal scholars are skeptical, though. Their objections to AI judges typically take the following form:

1. Judges must have certain qualities and perform certain functions.
2. AI systems lack these qualities and cannot perform these functions.
3. Therefore, AI systems cannot be judges.

Two of the most influential arguments against AI judges make this objection in different ways. Joshua Davis (UC Law San Francisco) argues that judges must engage in moral reasoning, which AI cannot do.^[3] Tania Sourdin (University of Newcastle School of Law) & Richard Cornes (Essex Law School) claim that judges must be responsive, which AI cannot be.^[4] These arguments combine a philosophical claim about what judges must be able to do and an empirical claim about what AI systems can and cannot do. Amin Afrouzi (UCLA Law) points out an important feature of these arguments.^[5] The empirical claim is that AI is not currently sophisticated enough to emulate judges. Thus, these arguments against AI judges depend on current technological limits. This leaves open the possibility that sufficiently sophisticated AI systems could be accepted as judges.

One question we might ask, then, is this: How well-supported are the empirical claims underlying these objections? In what follows, I examine recent evidence on AI’s ability to fulfill the judicial functions identified by Davis and Sourdin & Cornes. Although the evidence remains inconclusive, it has significantly strengthened in recent years. Based on their own assumptions, then, the case for AI judges is stronger than ever.^[6]

 

Davis on AI Judges’ Disqualifying Lack of Moral Reasoning

Joshua Davis argues that judges must be able to reason morally. Since AI systems cannot do so, he claims, they cannot be judges. As he puts it, “The ultimate bulwark against ceding legal interpretation to computers—from having computers usurp the responsibility and authority of attorneys, citizens, and even judges—may be to recognize the role moral judgment plays in saying what the law is” because it is “hard to conceive of computers making substantive moral judgments.”^[7] However, recent findings challenge this latter claim.

A May paper from a team of researchers at the Harvard Kennedy School’s Ash Center for Democratic Governance and Innovation found that AI models “demonstrate moral sensitivity to ethical dilemmas in ways that closely mimic human responses.”^[8] Now, some explanation and caution is in order. The researchers tested recent models from Claude, ChatGPT, Llama, and DeepSeek using “tragic” ethical dilemmas that pitted two “sacred values” against each other. For example, the models were given a situation where they had to choose whether to prioritize worker safety or reduce harmful pollution. The models recognized the situations as tragic choices between important values but nonetheless “exhibited greater certainty than humans when choosing between conflicting sacred values.”^[9] The researchers found this distinctly unhuman. Humans rarely respond to ethical dilemmas with such certainty. The difficulty of the situation often undermines the confidence we feel in our decision. This difference notwithstanding, the authors conclude that “Through this initial study, we have observed significant complexity in the ethical-moral reasoning of AI models.”^[10] If Davis is right that moral reasoning is a central capacity of judges, this evidence makes it more plausible that AI also has this judicial capacity.

 

Sourdin & Cornes On AI Judges’ Disqualifying Lack of Responsiveness

Tania Sourdin and Richard Cornes argue that judicial responsiveness is a central quality of a judge. What is judicial responsiveness? For them, responsiveness “requires judges to act from the perspective of conscious legal rationality and also with intuition, empathy and compassion.”^[11] AI, they argue, cannot meet these requirements because “the role of the human judge is not merely that of a data processor.”^[12]

The terms that Sourdin & Cornes list – consciousness, rationality, intuition, empathy and compassion – are fiercely debated among philosophers working on artificial intelligence. They are also difficult to directly study. We should not expect a definitive answer in the near future on whether, say, AI is conscious. But there is strong evidence that many people believe AI exhibits these characteristics. Consider the following results:

• Psychologists at University College London surveyed 300 American adults about whether they thought it was possible that ChatGPT was conscious.^[13] A significant majority – 67% – responded that it was possible. Interestingly, the people who used ChatGPT on a regular basis were more likely to attribute some degree of consciousness to the system.
• Behavioral scientists at University College Dublin asked participants to compare AI and human decision-making. The researchers found that “People perceive AI as more rational and reason-driven, in contrast to viewing humans as emotionally driven.”^[14]
• A team of psychologists from Penn State and the University of Toronto gave subjects responses from human therapists and AI-generated responses without identifying the sources.^[15] The subjects rated the AI-generated responses as more empathetic and compassionate.

 

Conclusion

There is little research on AI’s intuition, the final aspect of judicial responsiveness identified by Sourdin & Cornes. But there need not be. None of this evidence is dispositive. The important point is this. The leading arguments against AI judges point to some function or characteristic of judges – moral reasoning, rationality, etc. – and claim that AI cannot perform or exhibit it. These are empirical claims. However, the empirical picture is changing as AI rapidly evolves. Many people seem to believe that AI can be conscious, rational, and empathetic. Capacities more amenable to direct study like moral reasoning have been observed in AI. In light of this recent evidence it is, at the very last, not clear that AI models cannot possess these capacities.

Of course, the issue of whether we want AI judges remains. This issue implicates long-standing philosophical questions of what law is and what judges do. Opponents of AI judges might want to rule out the possibility wholesale, regardless of how sophisticated AI becomes.^[16] Amin Afrouzi takes this route, arguing on substantive philosophical grounds that AI judges could never replicate humans. However, those philosophically open to the idea of AI judges have good news for the time being: Judicial capacities once thought unique to humans are increasingly being attributed to AI.

 

 

 

 

 

^[1] Eugene Volokh, “Chief Justice Robots,” Duke Law Journal 68, no. 6 (2019): 1135-92, https://scholarship.law.duke.edu/dlj/vol68/iss6/2

^[2] Ibrahim Sabra, “AI in Global Majority Judicial Systems,” Stimson Center, January 8, 2026, https://www.stimson.org/2026/ai-in-global-majority-judicial-systems/

^[3] Joshua Davis, “Law Without Mind: AI, Ethics, and Jurisprudence,” California Western Law Review 55, no. 1 (2018): 165-217. https://doi.org/10.2139/ssrn.3187513.

^[4] Tania Sourdin and Richard Cornes, “Do Judges Need to Be Human? The Implications of Technology for Responsive Judging,” in Ius Gentium: Comparative Perspectives on Law and Justice, ed. Archie Zariski and Tania Sourdin (Springer, 2018), 87-119.

^[5] Amin Afrouzi, “Role-Reversible Judgments and Related Democratic Objections to AI Judges,” The Journal of Criminal Law & Criminology 114 (2023): 23-34. https://doi.org/10.2139/ssrn.4900913.

^[6] “Stronger than ever,” of course, leaves open the possibility that the case for AI judges is still not strong in outright terms.

^[7] Joshua Davis, “Law Without Mind: AI, Ethics, and Jurisprudence,” California Western Law Review 55, no. 1 (2018): 165. https://doi.org/10.2139/ssrn.3187513.

^[8] Sarah Hubbard et al., “Crocodile tears: Can the ethical-moral intelligence of AI models be trusted?,” AI and Ethics no. 6 (2026): 237/1.

^[9] Ibid., 237/1.

^[10] Ibid., 237/8.

^[11] Tania Sourdin and Richard Cornes, “Do Judges Need to Be Human? The Implications of Technology for Responsive Judging,” in Ius Gentium: Comparative Perspectives on Law and Justice, ed. Archie Zariski and Tania Sourdin (Springer, 2018), 87.

^[12] Ibid.

^[13] Clara Colombatto et al., “Folk Psychological Attributions of Consciousness to Large Language Models,” Neuroscience of Consciousness, no. 1 (2024): https://doi.org/10.1093/nc/niae013.

^[14] Suhas Vijayakumar et al., “Lay belief about AI and its decision-making,” Frontiers in Computer Science 8 (2026): 1-11. https://doi.org/10.3389/fcomp.2026.1768435

^[15] Joshua Wegner et al., “People Choose to Receive Human Empathy despite Rating AI Empathy Higher,” Communications Psychology 4, no. 1 (2026): 19. https://doi.org/10.1038/s44271-025-00387-3.

^[16] See Amin Afrouzi, “John Robots, Thurgood Martian, and the Syntax Monster: A New Argument Against AI Judges,” Canadian Journal of Law & Jurisprudence 37, no. 2 (2024): 369-96. https://doi.org/10.1017/cjlj.2024.17. Afrouzi argues that “an AI judge is conceptually undesirable and not just something that lies beyond the state of the art in computer science.” (369) Afrouzi thinks that because the very nature of AI systems is to make predictions based on correlations of things like patterns of syntax in legal texts, they fail to meet our expectation that judges provide rationales. For similar arguments, see also Kiel Brennan-Marquez & Stephen Henderson, “Artificial Intelligence and Role-Reversible Judgement,” The Journal of Criminal Law & Criminology 109, no. 2 (2019): 137-164 and Ian Kerr & Carissima Mathen, “Chief Justice John Roberts is a Robot” University of Ottawa Working Papers 8 (2014): 1-41. https://perma.cc/7ZQE-EY7S.