The ethics of AGI in judicial sentencing

"Image synthesis assisted by GPT Image 2.0, an AI partner within the Global Future Nexus ecosystem."

From risk assessment algorithms that predict recidivism to AI systems that generate sentencing recommendations, artificial intelligence is slowly entering the criminal courtroom. Yet the integration of AGI into sentencing raises profound ethical questions about justice, transparency, and the very nature of punishment.

The Promise of Algorithmic Justice

Proponents of AI-assisted sentencing see significant potential benefits . Algorithmic tools could reduce the influence of subjective and extra-legal factors—such as unconscious bias, judicial fatigue, and demographic disparities—that have long plagued sentencing. AI systems offer consistency: they are not swayed by emotion or the time of day, and they are capable of processing far more case data than any human judge. For supporters, AI sentencing is a path toward a cheaper, fairer, and more accurate criminal justice system.

In some jurisdictions, the first steps have already been taken. States have begun using algorithmic systems to provide sentence recommendations in serious cases such as drug possession and rape. The concept of “intelligent courts,” driven mainly by algorithmic systems, has even been declared an ambition by some governments.

The Ethical Terrain

Despite these promises, the scholarly debate has identified at least thirteen distinct ethical problems that AI sentencing raises. Some relate to the outcome of sentencing: Can an algorithm truly assess the severity of a crime? Does it produce accurate recommendations? Is it fair and equitable across different groups?

Others concern the process of sentencing. The “black box” problem—where the complex workings of machine learning models make their decisions opaque—raises serious concerns about transparency and the ability of defendants to challenge an AI’s assessment. The landmark case of Wisconsin v. Loomis highlighted this: the defendant was unable to scrutinise the reasoning behind the COMPAS risk assessment tool used in his sentencing, as its algorithm was protected as a trade secret. This lack of transparency can undermine the very legitimacy of the judicial process.

The Human Factor

The response of the judiciary itself is a critical factor. A study of Israeli district court judges identified two broad orientations. The first, held by “techno-optimists,” views sentencing as a normative reasoning process that could, in theory, be amenable to automation. The second, held by “techno-pessimists,” conceives of sentencing as a deeply phenomenological process that requires interpersonal-emotional engagement and moral deliberation—elements that cannot be reduced to algorithmic analysis. For these judges, the integration of AI into sentencing represents a threat to judicial professional identity and legitimacy. The resistance to “robot judges” is not merely technical; it is a defence of the human meaning of justice.

The GFN Context: Governance and Human Potential

For Global Future Nexus, the ethical governance of AGI in sentencing is a critical frontier. GFN’s Code of Ethics, with its commitments to Integrity Without Exception and Anticipatory Foresight, provides the framework for ensuring that AGI serves justice, not undermines it. The AI Identity Committee’s work on AGI subjective experience and role in society must extend to its role in the courtroom.

The path forward lies not in fully automated justice, but in a hybrid model. This requires the development of transparent algorithms, the creation of a new discipline of “Machine Psychology” to understand and mitigate emergent failures, and the establishment of robust oversight mechanisms that preserve human judgment as the ultimate authority. By ensuring AGI serves justice and human dignity, we can build a future where the integration of intelligence into the legal system strengthens the rule of law for all.

Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)

Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)