Regulating AI
M2 MVA · Semester 2 · Opening · 2.5 ECTS
Contents
- Introduction and overview
- What the module asks
- How to prepare
- Deliverables and assessment
- Official sources
Introduction and overview
| Module | Regulating AI |
| Official title | Regulating AI |
| Instructors | Erevan MALROUX (Mistral AI, per master-mva.com, 2026–2027) |
| Period | Semester 2 · period 2 (04/01/2027 – 02/04/2027) |
| Credits and volume | 2.5 ECTS · 12 h of lectures (4 sessions of 3 h) per the Université Paris-Saclay programme table and master-mva.com |
| Label | Opening · two Opening courses of 2.5 ECTS make one 5-ECTS module |
| Language | English (Université Paris-Saclay course detail) |
| Assessment | Prepared debate, 40 %: two groups of 3–4 students argue in a session (7 min per side, a 3-min reply), plus one common written summary of at most 4 pages; graded on completeness of arguments, anticipation of opposing arguments, clarity and concision — not on the opinion. Multiple-choice questionnaire on the course material, 60 %. |
| Prerequisites | None (Université Paris-Saclay course detail) |
| Other | Numerus clausus: 40 students |
This is a framing page, not a course: the module is a short law-and-policy seminar. The page says what the module asks, how a student of a mathematics and learning master can prepare for it, and how it is assessed.
The course description on master-mva.com sets three objectives. First, explore and categorise the risks that come with the development and deployment of AI, through a general typology. Second, understand and question how regulation can address them: its philosophy, purpose, content and limits. Third, study two European examples closely tied to AI systems: the data-protection regulation (GDPR) and the AI Act.
The Université Paris-Saclay detail lists four sessions: AI governance, a sectoral example, transversal laws (the EU AI Act and beyond), and foundation models.
What the module asks
- A typology of risks. Discriminatory bias, skewed representations of the world, lack of diversity, amplification of inequalities, privacy and surveillance, opacity of data and algorithms, diffuse responsibility. The course insists that these risks are systemic and cannot be left to regulators alone.
- Reading a regulation as a design. For the GDPR and the AI Act: who is regulated, what is required, how compliance is checked, and where the text leaves gaps.
- Arguing both sides. The prepared debate rewards completeness and anticipation of the opposing arguments, which is a scientific skill as much as a legal one.
How to prepare
The science pages of these notes give the technical side of most debate topics:
- fairness metrics, privacy and differential privacy: Responsible Machine Learning;
- explanation methods and their limits: Multimodal Explainable AI;
- foundation models, their training data and their evaluation: Foundations of Large Language Models and Training and Deploying Large-Scale Models;
- safety and alignment questions: Turing Seminar;
- environmental footprint of AI: Artificial Intelligence and the Environment.
A good preparation for a debate is a one-page table: the claim, the evidence for it, the strongest objection, and the answer to that objection.
Deliverables and assessment
- Debate (40 %): a 7-minute presentation per side, a 3-minute reply, and a common written summary of at most 4 pages.
- Multiple-choice questionnaire (60 %) on the course material.
These are the official rules from master-mva.com; this page adds no assessment of its own.
Official sources
- master-mva.com, course page “Regulating AI”, 2026–2027: https://www.master-mva.com/cours/regulating_ai/.
- Université Paris-Saclay, M2 Mathematics, Vision, Learning, programme table and course detail “Regulating AI”.
- Regulation (EU) 2016/679 (General Data Protection Regulation), Official Journal of the European Union, 2016.
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Official Journal of the European Union, 2024.