AI Works. But Do We Understand It? | Aude Billard on Robot Learning

In this episode of Creative Difference, Maxime Gabella speaks with Aude Billard, professor of robotics at EPFL, about a tension at the heart of modern AI: machines can increasingly perform remarkable tasks, yet we still struggle to explain how learning itself works. Starting from their shared background in physics, they explore the difference between making a system work and genuinely understanding it. What does a mathematical model reveal that a black-box neural network does not? Is scientific progress still progress when performance increases faster than explanation? Aude explains why apparently simple human abilities, such as grasping an object, placing an elbow, manipulating tiny watch components, or performing microsurgery, remain extraordinarily difficult for robots. These actions involve touch, force control, body awareness, prediction, adaptation, and the ability to isolate the few variables that matter in a complex environment. The conversation expands toward robot learning, transfer between different bodies, active inference, world models, foundation models, and the limits of data-driven AI. Can robots learn physical intelligence by observing vast amounts of human behavior, or must they act in the world and close the loop between perception and action? They also imagine robots that move beyond human anatomy: detachable crawling hands, machines that can walk, swim, fly, change shape, share information across bodies, and perceive aspects of reality inaccessible to human senses. Ultimately, the discussion returns to the central mystery: how do humans learn so efficiently, and what might building robots teach us about our own intelligence? The episode closes with a reflection on publication overload, scientific responsibility, creativity, education, and the need to slow down if we want to keep understanding what we create. Chapters 00:00 Introduction 00:36 Why Aude chose physics 03:11 Physics, philosophy, and the question of “why” 05:09 Publication overload and the loss of understanding 07:20 Are there fundamental laws of robot learning? 08:04 Watchmaking, touch, and embodied skill 14:48 Microsurgery and the efficiency of human learning 18:04 AI works, but do we understand it? 21:15 Augmenting humans and the possibility of a new species 23:38 Robots beyond the human body 26:49 New senses, shared world models, and imagination 33:01 Holistic robotics and the detachable hand 35:03 The mystery of how humans learn 38:36 Transferring knowledge between different robot bodies 44:01 Intuition, movement, and what words cannot explain 47:26 Active inference and world models 51:06 Foundation models versus closed-loop robotics 53:09 Why science must do more than “make it work” 55:39 Teaching robots by demonstration 01:00:18 Why science should slow down 01:02:27 Rethinking education and research #Robotics #ArtificialIntelligence #MachineLearning