The Use of Artificial Intelligence in Criminal Sentencing: Efficiency or Injustice?

By: April Pang Edited by: Isabella Canales and Chukwunonso Kojo Onwaeze

Introduction

From predictive policing to forensic analysis, generative AI  has begun to permeate every corner of the criminal justice system. Courts use AI to process documents, public defenders use it to draft motions, and judges increasingly rely on algorithmic risk scores when making sentencing decisions. For example, the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS), a risk assessment tool used in states such as Wisconsin, Florida, and New York to estimate and generate risk scores, can help simplify the judicial decision-making process and reduce human bias. [1] Supporters believe that these tools can enhance the efficiency of the system and even make it more equitable. Critics warn that AI could actually exacerbate injustice, undermine due process, and weaken the moral foundation of punishment. The central problem of this issue is no longer whether AI will be used for sentencing, but whether its use is justified.

The Case for Efficiency

AI has enhanced the efficiency of many links in the judicial system. In Palm Beach County, the accuracy rate of an automated document processing system is as high as 98% to 99%. This far exceeds the accuracy of manual operation, which enables staff to focus on other work that requires human discretion, such as exercising legal judgment or supporting litigants who need guidance. [2] The public defense attorneys' offices in Miami-Dade County and Los Angeles County use AI to organize case files by person rather than by case. This allows lawyers to build more humanized narratives and advocate for non-custodian alternatives to imprisonment. [2] These examples show that AI can reduce administrative burdens, which enables legal professionals to devote more time to tasks that require human judgment and rationally allocate their time and energy. AI‑assisted risk assessments also improve consistency. A Tulane University study of more than 50,000 Virginia cases found that algorithmic recommendations were sometimes corrected for judicial bias, even though judges often tended to ignore those recommendations for defendants of color. [2] Theoretically, standardized risk scores could reduce arbitrary disparities and help courts make more evidence‑based decisions.

When Efficiency Creates New Risks

While AI tools can process legal data with unprecedented speed and accuracy, their reliance on historical data introduces a much deeper systemic risk. As Villasenor and Foggo explain, the AI systems learn from historical data. The criminal justice data reflects decades of racial discrimination in policing and sentencing. [3] If the data is biased, then the output results of the algorithm will also be biased. Transparency is another major issue. In the case of State v. Loomis, the defendant questioned the use of the COMPAS risk assessment tool because he was unable to understand how the tool worked. The Wisconsin Supreme Court upheld the use of the tool, even though the design of the algorithm was protected by trade secrets. [3] This secrecy makes it difficult for defendants to question evidence against them and raises serious due-process issues.

Philosophers and computer scientists have also put forward different definitions of fairness, which conflict mathematically. As Wykstra points out, algorithms cannot simultaneously meet all fairness criteria, which means that any design choice involves ethical trade-offs. [4] These are not technical issues but rather value judgments about which definition of fairness the society should adopt and prioritize.

In addition to fairness, some scholars believe that algorithmic sentencing threatens the moral foundation of criminal punishment. Brennan-Marquez and Henderson hold that democratic legitimacy demands “role-reversible judgment"—that is, those who make the judgment must be able to be judged by others, and vice versa. [5] Algorithms cannot meet this requirement because they cannot be held morally or politically responsible. Similarly, Isaac Taylor argues that sentencing is an act of condemnation that expresses the moral judgment of society. Entrusting this function to algorithms may weaken the expressive value of punishment. [6]

A Threshold Model for Responsible Use

Given these tensions, we need a balanced framework. It is necessary to recognize the benefits of AI while setting clear limits on its risks. The threshold model offers a solution. Under this model, AI can only be used when the risk of it (especially racial discrimination) is below the set maximum level. This avoids the drawbacks of the purely utilitarian approach, as this method may justify unfairness in the name of efficiency. [7]

Transparency is the key to this model. Explainability Scores (xAI) can help quantify directly how understandable an algorithm’s reasoning is. Because many current systems achieve only 20–30% explainability, having a minimum threshold, such as 75% for general use and 80% for criminal sentencing, will ensure that the suggestions and conclusions made by AI can always be questioned and reviewed. [7] Regular audits help identify emerging biases and ensure that algorithms do not evolve into discriminatory patterns.

Maintaining Judicial Authority in an AI Era

AI should never be the final decision‑maker, even with supervision regulations. Judges and juries must reserve the final decision-making power. As Taylor argues, sentencing decisions must be placed under “meaningful public control” to maintain their punitive and democratic nature. [6] This means that humans, rather than algorithms, must bear moral responsibility for every judgment.

Public oversight is also essential. Many AI tools are developed by private companies, which has raised concerns about accountability and transparency. Transferring the supervisory power to public institutions can ensure that these tools reflect social values rather than commercial interests. As recent research on the application of artificial intelligence in the judicial system has shown, courts must establish a clear regulatory framework that requires regular audits and ensures that defendants can question algorithmic recommendations. [8]

Conclusion

AI has the potential to make criminal justice systems more efficient and humanized. However, without strong safeguard measures, it may also intensify prejudice and blur responsibility, as well as weaken the moral basis of sentencing. A responsible path forward requires clear acceptable risk thresholds, sound transparency standards, regular audits, and effective public supervision. AI should assist human judgment rather than replace it. Only in this way can efficiency and justice coexist in future criminal sentencing.

NOTES:

1. Kieran Newcomb, “The Place of Artificial Intelligence in Sentencing Decisions.”

2. Allyson Brunette, “Humanizing Justice.”

3. Villasenor and Foggo, “Artificial Intelligence, Due Process and Criminal Sentencing.”

4. Wykstra, “Philosopher1s Corner.”

5. Kiel Brennan-Marquez and Stephen E. Henderson, “ARTIFICIAL INTELLIGENCE AND ROLE-REVERSIBLE JUDGMENT.”

6. Alix Faulkner, “From Case Law to Code.”

7. Sienna Sedlarcik, “Ethical AI Sentencing: A Framework for Moral Judgment in Criminal Justice | Published in Critical Debates in Humanities, Science and Global Justice.”

8. Taylor, “Justice by Algorithm.”

BIBLIOGRAPHY:

Alix Faulkner. “From Case Law to Code: Evaluating AI’s Role in the Justice System.” Montreal AI Ethics Institute, May 26, 2025. https://montrealethics.ai/from-case-law-to-code-evaluating-ais-role-in-the-justice-system/

Allyson Brunette. “Humanizing Justice: The Transformational Impact of AI in Courts, from Filing to Sentencing.” Thomson Reuters Institute, October 25, 2024. https://www.thomsonreuters.com/en-us/posts/ai-in-courts/humanizing-justice/.

Kiel Brennan-Marquez and Stephen E. Henderson. “ARTIFICIAL INTELLIGENCE AND ROLE-REVERSIBLE JUDGMENT.” 2019. https://www-jstor-org.turing.library.northwestern.edu/stable/48572776?searchText=AI+in+criminal+justice&searchUri=%2Faction%2FdoBasicSearch%3FQuery%3DAI%2Bin%2Bcriminal%2Bjustice%26so%3Drel&ab_segments=0%2Fbasic_search_gsv2%2Fcontrol&refreqid=fastly-default%3A3a56d1b833cf636f767f0b59794ecb1a&seq=1.

Kieran Newcomb. “The Place of Artificial Intelligence in Sentencing Decisions.” Inquiry Journal, March 28, 2024. https://www.unh.edu/inquiryjournal/blog/2024/03/place-artificial-intelligence-sentencing-decisions.

Sienna Sedlarcik. “Ethical AI Sentencing: A Framework for Moral Judgment in Criminal Justice | Published in Critical Debates in Humanities, Science and Global Justice.” November 5, 2024. https://criticaldebateshsgj.scholasticahq.com/article/125464-ethical-ai-sentencing-a-framework-for-moral-judgment-in-criminal-justice.

Taylor, Isaac. “Justice by Algorithm: The Limits of AI in Criminal Sentencing.” Criminal Justice Ethics 42, no. 3 (2023): 193–213. https://doi.org/10.1080/0731129X.2023.2275967.

Villasenor, John, and Virginia Foggo. “Artificial Intelligence, Due Process and Criminal Sentencing.” Michigan State Law Review 2020, no. 2 (2020): 295–354.

Wykstra, Stephanie. “Philosopher1s Corner: What Is ‘Fair’? Algorithms in Criminal Justice.” Issues in Science and Technology 34, no. 3 (2018): 21–23.