A first-year law student comes across a tutorial on statistics applied to criminology. A lawyer specializing in personal data must understand how a scoring algorithm works to challenge an automated decision in court. These situations, once marginal a decade ago, are multiplying in law schools and firms.
Obligation to master AI: what the AI Act changes for legal professionals
Since February 2, 2025, Article 4 of the European regulation on artificial intelligence requires providers and deployers of AI systems to develop the mastery of AI among their staff. The obligation is proportional to the functions performed and the context of use.
Specifically, a notary is not asked to code in Python. They are asked to understand the capabilities, limitations, and risks of the tools they use daily. The nuance is significant, but it implies a minimal foundation of scientific culture that traditional legal training did not provide.
The AI Act is based on a classification by risk level. To determine the obligations applicable to a given tool, one must be able to assess its technical functioning, possible biases, and limitations. When exploring scientific subjects in law, it becomes clear that this skill transcends the traditional opposition between literary and scientific disciplines: scientific competence serves to correctly apply the legal rule.
Statistics and algorithms in daily legal practice
The most visible area is that of litigation related to automated decisions. When an employee contests a dismissal based on an algorithmic scoring, their lawyer must understand the logic of the model to identify potential discrimination.
AI systems that select candidates, assess skills, or detect prohibited behaviors during an exam are generally classified as high-risk systems. Their regulation requires skills in algorithms and evaluation that mere reading of the Labor Code does not provide.
What this implies for initial training
Several law schools are now integrating modules on descriptive statistics or formal logic into their undergraduate programs. We are not talking about fundamental mathematics courses, but about operational tools:
- Reading and interpreting statistical tables used in judicial expertise reports to avoid misunderstandings about correlations and causalities
- Concepts of probability applied to evidence, particularly in cases where DNA or fingerprints are involved
- Understanding the basic mechanisms of machine learning, sufficient to engage in dialogue with a technical expert during a dispute
Knowing how to read a technical expertise report is no longer optional for a legal professional. Feedback varies on the necessary intensity of these teachings, but the principle of a minimal scientific acculturation is no longer debated in professional circles.
Environmental law and scientific expertise: a case study
Environmental law has long illustrated the permeability between legal norms and scientific productions. When a judge must rule on the legality of an industrial project, they rely on impact studies, the reading of which requires knowledge in ecology, chemistry, or hydrology.
The issue is not that the magistrate becomes a biologist. It is that they can evaluate the reliability of an expertise without blindly depending on the discourse of the expert mandated by one of the parties. This capacity for critical judgment on scientific data is precisely what the scientific methodology modules in legal curricula provide.
The trap of scientific ignorance in judicial decisions
A judge who does not understand the difference between a statistical correlation and a causal link may validate a fundamentally biased expert reasoning. Such errors have concrete consequences for litigants.
Research on the relationship between legal norms and scientific productions shows that ignorance in scientific matters can be produced by the institutional process itself. Regulatory risk assessment sometimes generates its own blind spots due to insufficient skills among the legal actors involved.
Expected scientific skills according to legal domain
Not all legal professionals need the same background. A tax lawyer manipulates econometric models. A digital law specialist must understand how neural networks function. A criminal lawyer relies on forensic science and genetics.
| Legal Domain | Scientific Skill Mobilized |
|---|---|
| Digital Law and AI | Algorithms, machine learning, statistical biases |
| Environmental Law | Ecology, chemistry, reading impact studies |
| Criminal Law | Forensic science, genetics, probabilities (DNA) |
| Tax Law | Econometrics, financial modeling |
| Health Law | Pharmacology, epidemiology, clinical trials |
This table does not claim to cover all situations, but it shows that the intersection of science and law is neither a myth nor an exception. It is structural, and it is intensifying with the digitization of professional practices.
Continuing education for legal professionals in the face of AI
Article 4 of the AI Act does not only concern initial training. Lawyers, judges, and in-house legal professionals are also targeted. The measure requires that skills be proportionate to the tools actually deployed.
For a firm using an AI-enhanced case law research tool, this means that both partners and associates must understand the tool’s limitations: risks of hallucination, corpus bias, reliability scope of generated responses.
The scientific training of legal professionals has become a regulatory obligation, not just a competitive advantage. Professionals who still consider scientific subjects as foreign to their practice expose themselves to a double risk: that of misapplying the legal rule and that of failing to meet the competence requirements imposed by the European framework.
