Synthetic Lay Juries: Delegating Criminal Judgment to Artificial Intelligence?

The verdict in a courtroom is the sanctuary of human empathy. But what happens when the pursuit of efficiency pushes Artificial Intelligence into the heart of t

The criminal trial, with its solemn courtrooms, passionate arguments, and the crushing weight of the final decision, has always been considered the sanctuary of human fallibility, but also of human empathy. The lay jury represents the embodiment of this principle: ordinary citizens, called to judge their peers, weighing the evidence not only with the rigor of logic, but with the irreplaceable measure of life experience. Today, however, the insatiable drive toward efficiency is pushing Artificial Intelligence well beyond administrative support, to the point of grazing the very heart of the judicial system.

We are entering the era of experiments with "synthetic juries" and algorithmic judgment. Although there is not yet any court in the world where an AI formally sits in the jury box to decide on an individual's freedom, the academic and legal literature of 2026 demonstrates that this is no longer pure speculation. Algorithms are already used pervasively to select, profile, and exclude human jurors. But the next step is even more radical: placing an artificial intelligence model alongside citizens and magistrates to suggest verdicts of guilt or acquittal.

In this in-depth analysis for the Scenarios and Reflections column, we will explore the technological foundations and the enormous ethical criticalities of AI's entry into courtrooms. The thesis we will advance, supported by research in philosophy of law, is as uncomfortable as it is inescapable: the real problem does not lie in the potential, and debated, analytical superiority of the algorithm over the human mind. The critical knot is understanding whether, as a democratic society, we are willing to accept that a part of the moral judgment about freedom and punishment be delegated to an opaque system that, by its very nature, can never be called to answer morally for its own decisions.

1. From Profiling to Counsel: How AI Enters the Courtroom

To avoid slipping into dystopian sensationalism, it is essential to distinguish the current levels of AI infiltration into the criminal process, separating today's reality from future hypotheses.

  • AI as investigative and informational support: This is the level already widely in use. Algorithmic systems help courts collect and organize evidence, extracting data from smartphones, cameras, or financial records in times incompatible with human labor.
  • Algorithmic jury selection (Voir Dire): This is the current battleground. As documented by the University of Washington Journal of Law, Technology & Arts, AI is already used to mine public data and social media of potential jurors, applying predictive scoring models to suggest to prosecution and defense whom to exclude.
  • AI as decision-making consultant (The augmented judge): Controversial initiatives in which magistrates use risk assessment algorithms to decide the amount of bail or the length of sentences, based on predictive calculations of recidivism.
  • The hybrid and synthetic jury: This is the most radical scenario, currently the subject of study in papers such as "I, for One, Welcome Our New AI Jurors" published in the International Journal of Law. The idea ranges from placing a generative AI system (e.g., an LLM trained on penal codes) alongside a human lay jury during deliberations, to the extreme hypothesis of a panel composed entirely of AI agents to handle the caseload of minor offenses.

Juror profiling already represents an emergency. Analyses published in PMc Online warn that using AI to identify "biases" in citizens risks introducing algorithmic biases that are far more subtle. These profiling tools threaten to circumvent historic legal protections against discrimination (such as the Batson rule in the USA), codifying racial or class prejudices in a mathematical and, for this reason, apparently neutral manner.

2. The Algorithmic Pilot: How Do Judges and Jurors React?

What happens when an artificial intelligence decides a conviction? A recent study published in PLoS ONE ("Judges versus Artificial Intelligence in Juror Decision-Making") conducted experiments simulating the behavior of jurors who received verdict recommendations from a human judge or from an AI. The result is shocking for those who hoped for a total rejection of the machine: participants did not automatically avoid algorithmic judgment. In numerous contexts, they weighed the AI's opinion exactly on par with that of a flesh-and-blood judge, demonstrating a frightening readiness to delegate moral authority to software.

Even more revealing is the pilot study published in Taylor & Francis that compared real burglary sentences with those generated by large language models (such as Claude and Gemini). Surprisingly, the models reproduced with good approximation the reasoning of human magistrates. This apparent success is often used by technologists to promote the use of AI in support of the process, but it hides an enormous psychological risk: automation bias (the human tendency to blindly accept the machine's answer, considering it more "objective"). If the AI suggests a five-year sentence with formally impeccable logic, how many jurors or exhausted judges will have the energy and courage to dissent?

3. The "Fairness Gap" and the Perception of Legitimacy

Despite the temptation of efficiency, the social acceptance of these systems is profoundly problematic. A key concept that has emerged from research (see the Harvard Journal of Law & Technology) is the "human-AI fairness gap".

Studies on public perceptions, such as those analyzed in PMC and Springer, reveal that citizens perceive trials conducted entirely by human judges as significantly fairer, juster, and more legitimate. The population accepts, albeit with reservations, that the algorithm be used in the information and evidence gathering phase, but categorically rejects the machine meddling in the final decision, that is, the implementation of the sentence.

Curiously, these studies reveal unexpected demographic nuances. Some minorities, who have historically suffered injustice and prejudice from the prison system, sometimes show greater trust (or desperate hope) in AI-augmented decisions, clinging to the illusion that a machine, being devoid of emotions, can finally be purged of the systemic racism of human courts. Unfortunately, as a fundamental essay from Oxford Academic ("Iudicium ex Machinae") demonstrates, this hope is fallacious: the data on which AI trains contains centuries of inequalities, and the machine does nothing but learn them, amplify them, and hide them under a veneer of mathematical objectivity.

4. The Black Box of Moral Judgment: Why Opacity Is Unacceptable

A verdict in a courtroom is not a pure logical equation. It is an act of State power over an individual. For this reason, the legitimacy of the entire judicial system rests not only on the statistical "accuracy" of the sentence, but on three fundamental pillars, also highlighted in recent university theses (CEU Thesis): transparency, accountability, and contestability.

Current Artificial Intelligence (based on Deep Learning) is a "black box." It is unable to provide a transparent causal explanation of how it weighed an alibi or the reliability of an eyewitness. As the Oxford Journal of Law and Philosophy points out, while it is true that hybrid procedures (where the human judge consults the AI but retains final authority) can mitigate the perception of injustice, the knot of contestability remains. How does the defense attorney cross-examine an algorithm on appeal? How does one examine the reasonable doubt of a language model?

Delegating a substantial part of the deliberation to an AI raises devastating philosophical implications. Adjudication requires compassion, understanding of context, and the human capacity to feel the weight and remorse of one's decision. As the AJEE concludes in its study on the irreparable biases of judicial AI, a totally impartial artificial intelligence is unattainable, but above all, its insertion risks irremediably eroding public trust in the rule of law.

Key Operational Takeaways (Takeaways for Legislators and Jurists)

  • Ban Decision-Making Opacity: Ministries of Justice must prohibit the use of black-box algorithms in any decision-making or sentencing recommendation phase. If a software cannot explain the exact logic (with code traceability) that led it to suggest a particular sentence, its use must be constitutionally inadmissible in criminal matters.
  • Severely Regulate Algorithmic Voir Dire: The use of AI systems to profile potential jurors must be subject to rigorous standards and mandatory disclosure. Defenses must know if the prosecution is using software to systematically discard candidates based on hidden correlations (which often mask class or gender discrimination).
  • Preserve the Human Monopoly on Moral Judgment: AI can and must continue to evolve as a support tool for forensic analysis and case management. But the deliberation phase, the jury room, must remain an exclusively biological space. Empathy, doubt, and the responsibility of judgment are not defects to be corrected with silicon, but the only guarantees of our humanity.

Conclusions: The Unbearable Lightness of Delegation

The idea of placing an artificial intelligence alongside a judge or a lay jury stems from an apparently noble intention: to defeat the prejudices, inefficiency, and fallibility of human judgment. The world's courts are clogged and often unjust, and the surgical precision of the algorithm presents itself as an irresistible technocratic panacea.

However, the illusion of the impartial machine shatters against the harsh reality of jurisprudence. The criminal trial is not a math problem to be solved, but a social drama to be faced. When the defendant stands up to hear the sentence that will decide the next twenty years of his life, he has the constitutional right, and even before that a human right, to look into the eyes of the people who are condemning him.

The ultimate question we must confront before letting servers into the halls of justice is inescapable: if an Artificial Intelligence, armed with its ocean of data, helps to convict or acquit a human being, who is truly responsible for that fate? The citizen-juror who rested on the machine's answer, the judge who validated the output, or the mathematical model that suggested the decision and which, devoid of conscience, will never spend a single sleepless night over the doubt of having ruined an innocent person?

Bibliographic References and Sources

Article by the Editorial Staff of La Bussola dell’IA