Predictive Policing and AI: The Ethics of Public Order between Efficiency and Bias
The idea of preventing crimes before they are committed is no longer science fiction: it is the reality of AI-based "Predictive Policing." But what is the price
The idea of arresting a criminal before they even commit a crime has fascinated science fiction literature for decades. Today, this concept has a technical name: Predictive Policing. Through the use of Artificial Intelligence and Machine Learning algorithms, police departments worldwide are analyzing massive volumes of data to predict where, when, and by whom the next crime might be committed.
From a purely statistical and logistical standpoint, the promised efficiency is staggering. However, when we delegate to a machine the power to direct suspicion toward a citizen, we enter a constitutional and ethical minefield. Algorithms are not neutral oracles: they learn from the past. And if the past of law enforcement has been marked by bias, AI will only automate and conceal it behind the alibi of mathematical objectivity.
In this in-depth analysis, we will explore the profound legal and social implications of AI applied to public order. Through studies from international universities and INTERPOL guidelines, we will analyze the risk of feedback loops, the opacity of models, and the indispensable need to defend the human right to contestation.
1. The Promised Efficiency and International Governance
Today's police forces must manage an enormous amount of digital data: closed-circuit cameras, phone records, financial transactions, and GPS data. Human analysis is no longer sufficient.
Faced with this complexity, international institutions are trying to establish an ethical perimeter. INTERPOL, in its document on Introduction to Responsible AI Innovation (Toolkit), emphasizes that Artificial Intelligence should be seen as an investigative support tool and not as an autonomous decision-maker. The required governance principles include transparency, proportionality in the use of technological force, and legal accountability.
This tension between technological innovation and public responsibility is well summarized by the specialized magazine Police1, which in its essay on Ethical AI in law enforcement urges government agencies to find a balance between the need to thwart threats and the duty to protect citizens' privacy in order to maintain public trust intact.
2. The Original Sin: Historical Biases and "Feedback Loops"
The structural problem of predictive policing lies not in the code, but in the data with which this code is fed.
As explained with crystal clarity by the glossary of the Rutgers AI Ethics Lab dedicated to Predictive Policing, algorithms are trained on historical arrest data. If in the past the police patrolled certain neighborhoods disproportionately (often inhabited by ethnic minorities or low-income classes), the arrest rate in those areas will be statistically higher, regardless of the actual overall crime rate. The algorithm will read this data and send patrols back to that same neighborhood. More patrols will lead to more arrests (even for minor offenses), confirming the algorithm in a vicious cycle known as a Feedback Loop.
This risk of structural discrimination has also been extensively documented by the EUCPN (European Crime Prevention Network) in its report on the risks and challenges of Artificial Intelligence and predictive policing in Europe.
The invisible discrimination produced by machines risks destroying the principle of equality before the law. We analyzed how to mitigate this impact in our investigation on Algorithmic Bias and Access to Justice: From Invisible Discrimination to Solutions for a Digital Due Process.
3. Algorithmic Opacity and Constitutional Risks
When the algorithm labels a citizen as "high risk" for recidivism, the system collides with the fundamental principles of criminal law: the presumption of innocence and the right to defense.
An illuminating legal reading published in The National Law School Journal (JHULR), entitled Algorithmic Justice or Bias: Legal Implications of Predictive Policing Algorithms in Criminal Justice, highlights the problem of the so-called "Black Box." Many predictive policing software programs are developed by private companies protected by trade secrets. If a judge or police officer bases a precautionary decision on an algorithmic output, but the defense cannot analyze or question the source code that produced that output, the fundamental right to a fair trial is denied.
The intersection of proprietary code and personal liberty is one of the most serious drifts of the technological era. We discussed this in detail in our special feature dedicated to The Ethical Dimension of AI in Judicial Surveillance Processes: Justice or Automated Prejudice?.
4. Mass Surveillance and the Defense of Autonomy
The next step after geolocated predictive policing (which maps high-risk locations) is large-scale biometric and behavioral recognition, which maps people's intentions in real time. The use of smart cameras capable of identifying individuals in a crowd or detecting behavioral anomalies raises specters of mass surveillance.
In this scenario, human responsibility becomes the only barrier against dystopia. No Artificial Intelligence system, no matter how sophisticated, should have the power to automatically trigger police action without genuine critical validation by a trained human operator (Human in the Loop).
To fully understand the technologies at play and the strategies for protecting personal data, we invite you to read our focus on Mass Surveillance and AI: How to Defend Yourself in a Hyper-Connected Society.
FAQ: Artificial Intelligence and Predictive Policing
1. What exactly is "Predictive Policing"? It is the use of mathematical models, historical data analysis, and machine learning algorithms to identify potential criminal activity (where and when it will occur) or potential offenders, allowing law enforcement to proactively allocate resources in the field.
2. What is meant by "Feedback Loop"? It is a vicious algorithmic cycle. If historical data tells the AI that a certain neighborhood is at high risk for crime, the police will send more officers to that neighborhood. Having more officers on site will statistically generate more arrests, even for petty crime. The new data will confirm the AI's prediction, prompting it to send even more officers in the future, overestimating the real risk of that area compared to others.
3. Is the use of facial recognition for predictive policing legal in Europe? With the entry into force of the European AI Act, the use of "real-time" remote biometric identification in public spaces by law enforcement is strictly prohibited, except in exceptional and limited circumstances (e.g., imminent terrorist threat or targeted search for kidnapping victims). Biometric classification systems based on political beliefs or race are totally banned.
4. What does it mean that an algorithm is a "Black Box"? It means that the internal processes through which the algorithm (particularly Deep Learning neural networks) reaches a given conclusion are so complex that they are incomprehensible even to its creators. This makes it impossible for a defense lawyer to question the algorithm to understand why their client was flagged as a suspect.
5. How can law enforcement mitigate AI biases? Through three fundamental steps: 1) Using "Fairness Metrics" during model training to balance historical data; 2) Subjecting the software to cyclical and independent audits (reviews); 3) Ensuring that the final decision is always made by a human officer, specifically trained to recognize the system's limitations and false correlations.
Conclusions: Order and the Law
The use of Artificial Intelligence in crime fighting is an inevitable evolution. Renouncing the computational power of technology in the face of increasingly sophisticated criminal networks would be an institutional surrender.
However, efficiency can never become a pretext for sacrificing civil justice. When the prediction of a crime becomes the sole compass for police action, there is a risk of punishing citizens not for their individual faults, but for the statistical correlations generated by their zip code or the color of their skin.
The real challenge of 2026 is not to build algorithms that never make mistakes – an engineering impossibility – but to build a judicial system mature enough to know how to doubt the machines it has created, guaranteeing every individual the sacred right to prove the machine wrong.
Bibliographic References and Sources
To ensure legal, ethical, and institutional rigor, this article drew upon the following primary sources:
- Governance, Ethics, and International Guidelines:
- Analysis of Biases and Risks of Predictive Policing:
- Legal and Constitutional Implications:
- The National Law School Journal (JHULR) – Algorithmic justice or bias: Legal implications of predictive policing algorithms in criminal justice. Link