AI is not neutral. This module presents its risks in a structured way, drawing on studies and reference works.
One recent synthesis is authoritative: the International AI Safety Report 2026, an international scientific report chaired by Yoshua Bengio (Turing Award winner), which consolidates the state of knowledge.
6.1Cognitive risk: the erosion of our abilities
Several converging studies warn about cognitive offloading: by delegating our mental tasks to AI, we risk letting our own skills wither.
- A MIT Media Lab study (Kosmyna et al., 2025), using electroencephalography on 54 participants, shows that those who write with an LLM show lower brain activity and perform worse once the assistance is removed: a "cognitive debt".
- Another study (Gerlich, 2025) of 666 participants links heavy AI use to a decline in critical thinking, mediated by cognitive offloading.
- The 2026 Bengio report cites a study in which clinicians' ability to detect tumours without assistance had dropped by 6% three months after AI was introduced.
This field of research is still young, but the signal is clear enough to be taken seriously.
6.2Risk to jobs and inequality
The International Monetary Fund (Cazzaniga et al., 2024) estimates that about 40% of jobs worldwide are exposed to AI, and up to 60% in advanced economies. Of these exposed jobs, about half could be complemented (productivity gains) and the other half threatened.
A specific feature of this wave: unlike past automation, it first affects skilled jobs, which creates a risk of widening inequality.
6.3Democratic and information risk
Generative AI greatly multiplies the ability to produce credible disinformation (texts, images, voices, deepfake videos) at very low cost. What is at stake: manipulation of opinion, electoral interference, erosion of trust in any image or recording. The very possibility of doubting the authenticity of any content weakens public debate.
6.4Social risks: bias, surveillance, dependence
- Bias and discrimination: a model reproduces and amplifies the biases of its training data (recruitment, credit, predictive justice).
- Surveillance: facial recognition, exploitation of personal data.
- Emotional dependence: the use of AI companions raises questions of mental health and isolation.
6.5Safety and alignment risks
In the longer term, there is the question of alignment: how can we ensure that highly autonomous and highly capable systems really pursue the goals intended by humans? This is the subject of a dedicated field of research (AI safety) and of lively debates about large-scale risks.
6.6Environmental risk
Training and using large models consumes a lot of electricity and water (cooling data centres), at a time when demand is exploding.
Going further: the library of risks
French-language books
Intelligence artificielle, intelligence humaine : la double énigme
Daniel Andler
The reference philosophical analysis (2023).
L'Intelligence artificielle ou l'enjeu du siècle
Éric Sadin
A political and anthropological critique.
L'intelligence artificielle n'existe pas
Luc Julia
An engineer's demystifying view.
De l'autre côté de la machine
Aurélie Jean
Algorithms and bias.
La fin de l'individu
Gaspard Koenig
AI and free will.
Quand la machine apprend
Yann Le Cun
The view of one of the fathers of deep learning.
La civilisation du poisson rouge
Bruno Patino
The attention economy.
Schizophrénie numérique
Anne Alombert
Donner un sens à l'intelligence artificielle
Villani report
The founding document of the French strategy (2018).
International books
Superintelligence
Nick Bostrom
The classic on existential risks (2014).
Human Compatible
Stuart Russell
On the alignment problem (2019).
The Alignment Problem
Brian Christian
2020.
Weapons of Math Destruction
Cathy O'Neil
On algorithmic bias (2016).
Atlas of AI
Kate Crawford
The material, human and ecological cost of AI (2021).
Power and Progress
Acemoglu & Johnson
Who benefits from technological revolutions (2023).
The Age of Surveillance Capitalism
Shoshana Zuboff
2019.
Nexus
Yuval Noah Harari
Information and power, from writing to AI (2024).
The Coming Wave
Mustafa Suleyman
On the risk of proliferation (2023).
Finished this module?
Your progress is saved on this device.