« One of the most influential AI researchers in the world» selon la MIT Review, chief AI scientist chez Meta, professeur à New-York University, prix Turing 2019 – Yann LeCun soutient que l’intelligence artificielle pourrait, un jour, acquérir la compréhension d’un”modèle du monde », correspondant à une certaine représentation de ce monde. Pour le français, la progression de l’intelligence dite artificielle passe par l’acquisition de ce sens commun ou intelligence générale, et la capacité de prendre des décisions et de se comporter en fonction de ce sens commun.
Ce sens commun pourrait s’acquérir durant une phase, aujourd’hui inédite, d’apprentissage de l’IA qui constitue pour Y. LeCun « l’essence” de l’intelligence : “In many ways common sense amounts to the ability to predict what’s going to happen next. This is the essence of intelligence,” says LeCun. That’s why he—and a few other researchers—have been using video clips to train their models. But existing machine-learning techniques required the models to predict exactly what is going to happen in the next frame and generate it pixel by pixel. Imagine you hold up a pen and let it go, LeCun says. Common sense tells you that the pen will fall, but not the exact position it will end up in. Predicting that would require crunching some tough physics equations. That’s why LeCun is now trying to train a neural network that can focus only on the relevant aspects of the world: predicting that the pen will fall but not exactly how. He sees this trained network as the equivalent of the world model that animals rely on » commente la MIT Review à ce sujet (1).
A la lecture de l’article, on comprend que de nombreuses questions restent en suspens : “LeCun is convinced that something like a configurator is needed, but he doesn’t know how to go about training a neural network to do the job. “We need to figure out a good recipe to make this work, and we don’t have that recipe yet,” he says. In LeCun’s vision, the world model and the configurator are two key pieces in a larger system, known as a cognitive architecture, that includes other neural networks—such as a perception model that senses the world and a model that uses rewards to motivate the AI to explore or curb its behavior. Each neural network is roughly analogous to parts of the brain, says LeCun. For example, the configurator and world model are meant to replicate functions of the prefrontal cortex. The motivation model corresponds to certain functions of the amygdala, and so on“.
“Yann LeCun has a bold new vision for the future of AI pulls together old ideas to sketch out a fresh path for AI, but raises as many questions as he answers » concluent les auteurs de l’article. Car en effet la question n’est pas seulement d’imaginer comment faire raisonner les robots comme des humains, mais aussi et surtout de savoir où se trouve l’intérêt des humains dans ce domaine.
0 commentaire