Intelligence Artificielle, bilan d’un engouement

Nicolas Courty, professor and head of the Obélix research team, delivers a conference on artificial intelligence, aiming to demystify the subject and provide a balanced view. He begins by defining AI as tasks currently better performed by humans, emphasizing the shift from symbolic AI to statistical learning. He explains the concept of a mathematical function and how neural networks are used to approximate complex functions from data. The success of modern AI is attributed to increased data availability and computational power (GPUs). He highlights beneficial applications such as medical diagnosis, drug discovery, and CO2 capture, as well as his team’s work on intelligent fishing nets. He then discusses generative AI, large language models, and the phenomenon of hallucination. He addresses the limitations of AI, including bias, lack of common sense, and the black-box nature. He explores the societal implications, including job displacement, misinformation, and the concentration of power. He concludes by emphasizing that AI is a tool whose impact depends on human choices, advocating for a third voice in the debate beyond techno-optimism and doom.