Introduction 

        Similar to growing trends throughout society, courts across the United States are following suit, with an increase in the use of algorithmic risk-assessment systems in their proceedings. An example of such is Correctional Offender Management Profiling for Alternative Sanctions (COMPAS), which is used to predict recidivism. Algorithmic risk-assessment systems advertise ideal objectivity and efficiency in court proceedings, but ultimately operate as opaque, proprietary algorithms. Before the use of these systems, judges customarily relied on discretion and presentence reports to make their decisions. Today, machine learning systems are used to determine sentencing, bail, and parole decisions. Issues with such systems arise because these algorithms are proprietary; the defendants in these hearings and cases cannot examine how their scores are generated, as the algorithms produce only answers without reasoning or justification. These algorithms provide insufficient evidence that their code lacks prejudice or discriminatory effects against particular defendants, raising concerns in the legal system about the fairness of due process for these defendants. Thus, although algorithmic risk assessment tools promote neutrality and efficiency in court, their design compromises the constitutional guarantee of procedural due process by preventing defendants from challenging the accuracy and justice of the information used to determine their liberty. 

 

Due Process and the Right to Fair Sentencing Procedures

        Extending the 5th Amendment Due Process Clause, the 14th Amendment obliges not only the federal but also the state governments to provide procedural due process to all residents, including non-citizens, within their respective jurisdictions [1]. Due process is a legal principle requiring government officials to follow established procedures [2]. Such that each defendant suspected of committing a crime is entitled to a notice and hearing before an impartial jury of their peers before depriving the defendant of life, liberty, or property [3]. Procedural due process becomes pertinent to the ethics and importance of using algorithms in court cases as it sets guidelines to ensure each defendant is ensured the same rights and privileges, removing any contingencies. 

        As in Mathews v. Eldridge, George Eldridge filed a lawsuit against the Department of Education and Welfare, the federal agency that, at the time, administered disability benefits, for the Social Security Administration's violation of his 14th Amendment rights by terminating his Social Security benefits without a proper hearing [4]. This case sets up the legal framework for procedural due process in relation to court algorithms. Eldridge’s case was won in the Federal District Court, the Court of Appeals, and the United States Supreme Court. Within each hearing, a three-tier risk-assessment algorithmic system was used to determine whether the government had discriminated against the defendant. In the first tier, the private interest is affected, meaning the Defendant’s liberty and the length of their incarceration are affected [5]. This could encompass the loss of a professional license, personal liberty, or, in Mr. Eldridge's case, welfare benefits. Secondly, the tier system evaluates the risk of error in the procedure used, as well as the case's evidence, personal appeal, and transparency [6]. Errors could come from biased data, secret evidence, or the inability to cross-examine. Reasonably, COMPAS raises concerns for this second tier because of its proprietary nature; the defendants cannot see how their scores are calculated because of the algorithm's coding. This process only provides a single output, which obscures the underlying logic used to generate the code. Without transparency into this decision-making, users lack the evidence necessary to challenge the ruling. The third and final tier is Government interest. During such rulings, the courts consider managerial efficiency, cost savings, and public safety [7]. Their arguments for using algorithm systems rely on the belief that they improve the consistency of court sentencing, reduce judicial bias, help protect public safety, and save time by decongesting court systems [8]. While algorithmic tools claim to reduce error, their lack of transparency may, in turn, increase it.

​

[1] U.S. Const. amend. XIV, § 1.

[2] U.S. Const. amend. XIV, § 1.

[3] Ty McDuffey, 14th Amendment Summary, LegalMatch (2023), https://www.legalmatch.com/law-library/article/14th-amendment-summary.html 

[4] Mathews v. Eldridge, 424 U.S. 319 (1976).

[5] Mathews v. Eldridge, 424 U.S. 319 (1976).

[6] Mathews v. Eldridge, 424 U.S. 319 (1976).

[7] Mathews v. Eldridge, 424 U.S. 319 (1976).

[8] Riccardo Fogliato et al., "Global and Local Interpretability in Criminal Justice," Harvard Data Science Review (2021), https://hdsr.mitpress.mit.edu/pub/hzwo7ax4/release/7. 

 

Algorithmic Sentencing in Practice: The Loomis Case 

        Many cases across the United States have experienced transparency errors created by algorithms. For instance, in State v. Loomis, the defendant Eric Loomis challenged the use of a COMPAS score in sentencing. Loomis was sentenced to six years in prison for fleeing a traffic officer and driving a vehicle without the owner's permission. Asserting a violation of his due process rights, Loomis challenged the use of COMPAS scoring on the grounds that the algorithm’s proprietary status shielded its decision-making process from examination [9].

        The court allowed the use of COMPAS but required judicial warnings. Consequently, algorithmic tools may not function as the exclusive basis for judicial determinations; rather, they must be subject to comprehensive traditional judicial review. COMPAS uses a 75-question test based on interviews, records, and self-reports to generate risk scores of low, medium, or high. Because COMPAS does not explain its reports, it cannot serve as the sole basis for sentencing. Judges must understand the system's limitations; thus, judicial warnings must be used as a precaution. Misusing this technology may result in bias and further violations of liberty.

        The controversial nature of State v. Loomis is that the defendants cannot effectively challenge evidence that they cannot inspect. The COMPAS algorithm's proprietary response lacks justifiable reasoning, which could be used to disprove a defendant's justification for being deprived of their liberty. In this way, the use of COMPAS violates the second tier of the 14th Amendment, which permits courts to accept secret evidence in sentencing. This, in turn, creates a direct conflict with earlier Supreme Court precedent on sentencing transparency.

​

[9] State v. Loomis, 881 N.W.2d 749 (Wis. 2016).

 

The Constitutional Problem: Secret Evidence in Sentencing 

        Garner v. Florida is another case exemplifying the misuse of algorithmic systems in court. The year 1979 marks when Robert Gardner was convicted of the murder of his wife in the case Gardner v. Florida [10]. Gardner was sentenced to death by a judge who used ‘secret information’, one of the pillars used in the justification of the 14th Amendment. The judge had ordered a presentence investigation report, which had key confidential information, that the judge had used in his decision to sentence Gardner to death despite other jurors recommending only life imprisonment [11]. The information used in Gardner’s sentencing was withheld from both Gardner and his attorney; neither was allowed to see it, and Gardner was therefore unable to dispute it, thereby violating his right to due process under the 14th Amendment. As such, the case was taken to the Supreme Court, where Gardner won with a 7-1 vote. In their decision, the court held that a defendant must have the liberty to review and challenge the evidence used against them in court, as capital sentencing requires heightened procedural fairness because the death penalty is irreversible [12]. Thus, the use of secret evidence violates due process as a defendant cannot be sentenced based on information they have no opportunity to deny or explain. The same reasoning applies to COMPAS algorithms. A defendant cannot challenge input variables, statistical assumptions, or data correctness, all of which would be considered secret information in law.

        The legal framework permitting the use of such algorithms is rooted in corporate trade secret protections. These programs allow technology companies to assert proprietary protection over the algorithms, while defendants claim constitutional discovery rights. The Corporate trade secret law protects proprietary business information such as formulas, customer lists, and processes [13]. These laws are security measures protected by both state and federal law. Yet algorithms can contain the potential to discriminate based on factors such as race and sex, which should not be protected under corporate trade secret law when used with algorithms. Ultimately, the legal system must reconcile intellectual property protection with the need for transparency to ensure that proprietary algorithms do not become a shield for systemic bias.

 

[10] Gardner v. Florida, 430 U.S. 349 (1977).

[11] Gardner v. Florida, 430 U.S. 349 (1977).

[12] Gardner v. Florida, 430 U.S. 349 (1977).

[13] Mitchell Williams (2024). What is the Definition of a Trade Secret Under Federal Law?, Williams Law Firm Blog. https://www.mitchellwilliamslaw.com/what-is-the-definition-of-a-trade-secret-under-federal-law.

 

Algorithmic Bias and Equal Protection 

         Algorithms used in court raise concerns when proxy variables and hidden biases are suspected to be in their code. Algorithms frequently exclude race but include variables like neighborhood, income, education, and arrest history that can function as racial proxies and thus discriminate regardless of the blatant racial bias coding that may be seen throughout court algorithms [14]. However, it must be duly noted that there is a difference between discriminatory effect and discriminatory intent. 

        Found in the 14th Amendment, the Equal Protection Clause requires that states treat individuals in similar situations equally under the law, thereby forbidding any and all discriminatory classifications based on race, gender, or other characteristics [15]. Though the Equal Protection law requires proof of intent. Issues arise, however, as algorithms may produce racially disparate outcomes free of explicit racial inputs, with the intent to discriminate. Because of this, the First Step Act must be referenced, as it requires federal risk assessments, such as COMPAS, to be statistically sound. Yet, some argue that this does not eliminate structural bias, as references to factors such as neighborhoods and education create racial proxies that inadvertently discriminate against defendants [16]. Thus, even in the absence of explicit discriminatory intent, the reliance on proxy variables in algorithmic sentencing raises serious Equal Protection concerns by perpetuating racially disparate outcomes that the law struggles to adequately address.

 

[14] Molly Glass. (2023) Algorithms Were Supposed to Reduce Bias in Criminal Justice – So They?, The Brink. https://www.bu.edu/articles/2023/do-algorithms-reduce-bias-in-criminal-justice/.

[15] Cornell Law School. (2023). Equal Protection, Legal Information Institute. https://www.law.cornell.edu/wex/equal_protection.

[16]  Federal Bureau of Prisons. (2023). First Step Act Overview. https://www.bop.gov/inmates/fsa/overview.jsp.

 

Toward Constitutional Oversight of Algorithmic Justice

        The complications arising from the use of AI algorithms in legal disputes necessitate change. Firstly, courts should ensure transparency in algorithms. Risk assessment tools used in court should be open to inspection by both defendants and experts. This would allow for any and all errors to be rectified before misuse of the systems can affect the court's ruling. Secondly, there should be evidentiary limits. Algorithms should primarily serve as supplementary information, never as primary evidence. And finally, courts should have independent auditing. Courts should require scheduled audits of bias and accuracy conducted by neutral researchers. These reforms would therefore allow courts to benefit from technological innovation without sacrificing constitutional protections. 

 

Conclusion 

        Current risk assessment algorithms pledge efficiency and objectivity. However, their overall opaque design undermines procedural fairness and violates the 14th Amendment. A justice system that relies on secret algorithms risks replacing judicial discretion with unaccountable technology. The Constitution requires that liberty decisions remain transparent, challengeable, and humanly accountable in the defendant's favor. If artificial intelligence is to play a role in sentencing, the law must guarantee that technological efficiency never supersedes fundamental constitutional rights.