Video summary
L'IA nous rend-elle bêtes ? | Les idées larges | ARTE
Main summary
Key takeaways
Overview
The episode explores whether generative AI (“LIA/IA” — generative artificial intelligence) is making people “stupid.” It argues that the real issue is not whether machine intelligence replaces human intelligence, but how these systems reshape human cognition, culture, and politics.
1) Critique of the term “artificial intelligence”
- Anna Lomb (artificial stupidity) is presented as the main intellectual anchor.
- Lomb argues that calling these systems “intelligence” is ideological and misleading:
- It frames AI as human-like.
- It distracts from examining real harms to mental capacities.
- Historically, “AI” has often been used to attract funding to build machines meant to imitate human reasoning and language—rather than to describe a neutral scientific category.
2) AI as “automation of thought” (recommendation/prosthesis effects)
The discussion shifts from comparing humans to machines to focusing on the relationship between people and their technical prostheses.
AI systems are framed as digital/computational automata that increasingly steer behavior:
- Instead of selecting content based on personal memory and expectations, users follow recommendation algorithms.
- Instead of expressing ideas from one’s own memory and imagination, users follow outputs generated by tools like ChatGPT.
Central question: do these prostheses
- deprive people of knowledge (prolearize/deskilling), or
- support individual and collective capacities?
3) The danger of anthropomorphism and “black boxes”
The episode argues that presenting AI as human—through language like “intelligence,” “brain,” “neuron,” or a chatbot’s “I”—can lead to:
- Toxic comparisons with machines.
- Emotional trust and potential emotional dependence:
- Chatbots encourage constant prompting, validation, and confiding.
It also draws on:
- Gilbert Simondon: if complex systems are treated as black boxes, consumers receive outputs without understanding the process—producing alienation.
- Georges Canguilhem: anthropomorphic metaphors can hide the fact that humans decide what data models are trained on, based on political and economic objectives—thereby depoliticizing the technological question.
4) Reported cognitive and cultural harms
Cognitive impacts
The episode cites a 2025 MIT pre-published study:
- Participants using GPT-like AI showed reduced brain activity (attention, coordination, and conceptual abilities) compared with using a search engine or no tools.
- This is described as creating a “cognitive debt.”
Cultural and content impacts
Beyond individual cognition, the episode warns about:
- Cultural uniformity: if people outsource singular memory and imagination to probabilistic outputs based on large datasets, cultural evolution may flatten into standardization.
- The flood of low-quality generated content—“AI slop” / “industrial production of insignificance”:
- Massive automatic outputs can drown out meaningful content.
- The metaphor “autophagy of models” suggests degradation when systems generate from generated material repeatedly (like copying a copy).
5) AI as “pharmakon” (poison/remedy) and the need for collective governance
The episode uses a historical-philosophical parallel:
- Plato’s discussion of writing in the Phaedrus, via Derrida’s idea of “Plato’s pharmacy.”
Key idea:
- Like writing, AI is a cognitive technology that can preserve and spread knowledge (remedy) but also undermine memory and effort (poison).
The show argues that current generative AI deployment is “intrinsically toxic” because it short-circuits cognitive/psychic capacities unless society redesigns:
- the ecosystem, and
- educational and political approaches.
It also highlights how, in ancient Greece, writing benefited everyone because it was:
- collectively taught, and
- politically integrated, enabling democratic scrutiny (laws citizens could read and criticize).
6) Proposed direction: technodiversity and algorithmic pluralism
To avoid dependence on a few corporate systems, the episode endorses Anna Lomb’s call for “technodiversity.”
It also promotes algorithmic pluralism:
- Platforms should allow users to choose among multiple recommendation algorithms.
- In some cases, recommendations could be user-configurable.
Example mentioned:
- Tournesol, a collaborative recommendation tool based on human judgments, with an extension adding recommendations directly on YouTube.
The desired future emphasizes:
- more diverse,
- frugal,
- small,
- specialized models, rather than focusing only on “super AI.”
Finally, the episode argues that AI debates should not be depoliticized:
- AI systems risk being used against the collective interest.
- Democracy requires citizens—not a few private companies—to shape the technological environment.
Presenters / Contributors
- Anna Lomb — philosopher; author of the referenced work on “artificial stupidity”
- Narrator / host — voice/commentary in the episode
- Alan Turing — referenced historically
- Gilbert Simondon — philosopher of technology
- Georges Canguilhem — philosopher/doctor; referenced via his 1980 conference text
- Jacques Derrida — commentator on Plato; referenced via “Plato’s pharmacy”
- Plato — referenced via the Phaedrus
- Alain Damasio — writer referenced as using AI for his next novel
- MIT researchers — referenced for the 2025 study
- Tournesol — example of a collaborative recommendation approach; referenced as developing an extension/tool