General AI, artificial intelligence that could surpass humans, is the subject of much discussion. It even seems that it is for this year: But do we have a common definition of what it could be?

With

• Jean-Gabriel Ganascia: Professor of computer science at the Faculty of Sciences of Sorbonne University and senior member of the Institut Universitaire de France
• Giada Pistilli: Philosopher specializing in the ethics of artificial intelligence and AI ethicist at Hugging Face
• Axel Cypel: Mining engineer, author and speaker in the field of AI

Paris is preparing to host the Summit for Action on Artificial Intelligence. As for the tech giants, they aim to make a technological leap and move from so-called weak AI to so-called general AI. But do we have a consensual definition?

This more ambitious form of AI aims to surpass the performance of the human brain: an autonomous super-intelligence, capable of adapting to everyone. The idea is in embryo in all the major research laboratories. According to Sam Altman, the boss of Open A.I, it even seems that it could be for this year. While Donald Trump has announced injecting 500 billion dollars into the sector and China has entered the race, and on the stock market, is the advent of strong AI scientifically plausible? And are the concerns it raises founded?

How to use language models in robotics?

A strong artificial intelligence capable of doing everything like a human should necessarily be able to make everyday gestures, like opening a door or grabbing a glass of water. Therefore, it needs a physical body, which means that it must be coupled with a robot. Linking robotics with the latest advances in artificial intelligence is not an easy task, but the Machine Learning and Information Access team at the Institute of Intelligent Systems and Robotics in Paris is trying to meet this challenge: Laure Soulier, lecturer at Sorbonne University, and her doctoral student Salim Aissi, explain how they hybridized a large language model – the type of model that Chat GPT made famous – with a reinforcement learning model, more often used in robotics.