When the Algorithm Takes the Bench: AI, Cognitive Bias, and the Future of Judicial Legitimacy

How artificial intelligence is quietly reshaping who — and what — decides justice

Imagine standing before a judge — not a person in a robe, but a statistical model trained on decades of past verdicts. The algorithm has already assessed your risk score, flagged your demographic profile, and generated a sentencing recommendation. The human judge reads it. How much does it influence the outcome? More than we’d like to think.

A new academic paper published in the International Journal of Social Sciences asks a question that should concern anyone who cares about justice: What happens to judicial legitimacy when artificial intelligence enters the courtroom?

The article, “Reconfiguring Judicial Legitimacy: Artificial Intelligence, Cognitive Bias, and the Rise of Algorithmic Authority” by Assoc. Prof. Dr. Fikret Erkan, argues that we are witnessing a quiet but profound transformation in how judicial authority is constituted — and that without careful safeguards, the rule of law itself may be at risk.

The Hidden Bias of Human Judges

Before we can understand the risks of AI in courts, we need to acknowledge something uncomfortable: human judges are already biased.

Erkan’s paper draws extensively on cognitive psychology to catalog the types of bias that routinely affect judicial reasoning. These include:

Anchoring bias — where initial information (like a prosecutor’s sentencing request) disproportionately influences the final outcome. Confirmation bias — where judges unconsciously seek information that supports their initial impressions. Attribution bias — which affects how judges interpret the motivations of defendants, often in ways that disadvantage those from marginalized groups. In-group bias — where judges show more lenient treatment toward defendants who share their social or cultural background.

These are not fringe findings. Decades of behavioral research confirm that judicial decisions are influenced by factors that have nothing to do with the law: the time of day, a judge’s mood after lunch, the race and gender of the defendant, even the order in which cases are heard.

Erkan makes a crucial point: judicial impartiality was never a natural attribute of adjudication. It has always been a fragile institutional ideal that requires active protection. The problem is that AI doesn’t solve this problem — it transforms and potentially amplifies it.

The Algorithm Is Not Neutral

The paper’s most important contribution may be its analysis of algorithmic bias — and why calling AI “objective” is itself a dangerous myth.

Algorithmic systems used in courts are trained on historical data. That data reflects decades of human decisions — including all their biases. A risk assessment tool trained on past recidivism rates will inevitably encode the racial and socioeconomic disparities that existed in those past decisions. The algorithm doesn’t create bias from scratch; it inherits, systematizes, and scales it.

Erkan identifies three key sources of algorithmic bias in judicial contexts:

Training data bias — when historical patterns of discrimination are baked into the model’s foundations. Architectural bias — the design choices and optimization objectives of the model itself, which reflect the values of those who built it. Feedback loops — when biased outputs feed back into future training data, compounding errors over time.

Perhaps most troubling is the “black box” problem. Many AI systems cannot explain why they reached a particular conclusion. This directly conflicts with a foundational principle of constitutional law: the right to a reasoned decision. If a judge cannot explain — or does not know — why an algorithm flagged someone as high-risk, how can that decision be challenged on appeal?

The paper also addresses the automation bias phenomenon: the documented tendency of human decision-makers to defer to algorithmic outputs, especially under cognitive load or time pressure. The very presence of an AI recommendation can crowd out independent judgment.

The Legitimacy Crisis: When Statistics Replace Justice

At the heart of Erkan’s paper is a theoretical claim that goes beyond individual bias cases: the rise of AI in courts represents a transformation in the very source of judicial authority.

When a judge cites case law, constitutional principles, and legal reasoning, their decision derives legitimacy from what Max Weber called “rational-legal authority” — the idea that power is legitimate when it follows established rules and procedures. But when a judge defers to a statistical prediction, the basis of legitimacy shifts from normative reasoning to probabilistic accuracy.

This is more than a philosophical distinction. It has concrete constitutional consequences. Consider:

If an algorithm predicts someone is 73% likely to reoffend, and a judge uses that to justify detention, the defendant is essentially being punished for a future crime they haven’t committed. The algorithm is aggregating patterns from thousands of other people’s histories and applying them to an individual case. Statistical probability is not the same as individual culpability.

Erkan argues that this creates a fundamental conflict between the logic of algorithms and the logic of law. Algorithms optimize for accuracy across populations. Courts adjudicate the rights of individuals. These are not the same thing — and conflating them erodes the constitutional foundations of the justice system.

A Four-Layer Solution: The ECASM Model

So what is to be done? Erkan’s answer is the Erkan Constitutional Algorithmic Safeguard Model (ECASM) — a four-layer framework for integrating AI into judicial processes without sacrificing constitutional legitimacy.

Layer 1: Normative Compatibility Review. Before any AI system can be deployed in a judicial context, it must undergo rigorous review to ensure it is compatible with constitutional rights — particularly equality, due process, and non-discrimination. This means evaluating not just the algorithm’s technical performance but its normative assumptions.

Layer 2: Transparency and Explainability Requirements. AI systems used in courts must be explainable. Judges, defendants, and appellate courts must be able to understand how a recommendation was generated, what factors were weighted, and why. This is a constitutional requirement, not just a technical preference.

Layer 3: Institutional Accountability Safeguards. There must be clear chains of accountability. Who is responsible when an algorithm contributes to an unjust outcome? Courts, developers, government agencies? The ECASM requires institutional structures that assign responsibility and create mechanisms for redress.

Layer 4: Behavioral Integrity Controls. This layer focuses on human behavior — specifically, on preventing automation bias. Judges must be trained to use AI as a tool, not a crutch. Protocols should be designed to ensure that algorithmic recommendations are considered alongside, not instead of, independent legal reasoning.

What the Courts of the World Are Actually Doing

The paper situates its analysis in a comparative legal context. Across the United States, Europe, and beyond, courts are already deploying AI tools — with wildly varying degrees of oversight and accountability.

In the United States, tools like COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) have been used in sentencing decisions for years, despite being commercially proprietary and essentially opaque to defendants. The Wisconsin Supreme Court case State v. Loomis (2016) upheld the use of COMPAS risk scores while noting concerns about transparency.

In Europe, the EU AI Act (2024) represents the most comprehensive regulatory framework to date, classifying AI systems used in the administration of justice as “high-risk” and imposing strict requirements for transparency, human oversight, and technical documentation. However, implementation and enforcement remain ongoing challenges.

Erkan argues that existing regulatory frameworks, while encouraging, do not yet constitute the kind of constitutionally grounded multi-layer oversight that the ECASM proposes. Technical compliance is not the same as normative legitimacy.

The Bottom Line: Algorithms Can Assist Justice, But Cannot Embody It

Erkan’s paper closes with a formulation that deserves to be quoted and repeated: in the AI era, judicial impartiality cannot be reduced either to human virtue or to algorithmic accuracy. It must be reconstructed as a dynamic constitutional equilibrium between human normative responsibility and technologically mediated decision-making.

This is not technophobia. The paper does not call for banning AI from courts. It calls for something harder: subjecting AI to the normative framework of law, rather than allowing law to be reshaped by the logic of algorithms.

For those of us who care about plain language and accessibility in law, this paper raises some urgent questions:

When an algorithm affects someone’s liberty, do they have the right to understand how it reached its conclusion? When AI recommendations are embedded in court decisions, how do defendants — especially those without legal expertise — challenge them? Who bears responsibility when an opaque algorithm contributes to an unjust outcome?

These are not purely technical or legal questions. They are questions about power, accountability, and what kind of justice system we want to build. The answer cannot be left to engineers and lawyers alone.

In the age of artificial intelligence, the rule of law demands more than code. It demands constitutional courage — the willingness to subject technological power to the norms of human dignity and equal treatment.

---

Source: Erkan, F. (2026). “Reconfiguring Judicial Legitimacy: Artificial Intelligence, Cognitive Bias, and The Rise of Algorithmic Authority.” International Journal of Social Sciences (IJSS), Vol. 10, Issue 42, pp. 159–220. DOI: 10.52096/usbd.10.42.09