“There are two different kinds of question we ask today. One question is: how do we generate new ideas of problems, goals, or games? And the second one is: how do we select which one to focus on, in priority, when we are in a situation where we can express our curiosity?” ~ Pierre-Yves Oudeyer

I’m a baby-besotted grandmother and — if such a thing is possible — that’s only increased my interest in how curiosity helps us learn.

How do we make anything of anything when we’re learning from scratch?

It turns out the answer lies in making that learning our own.

Researcher Pierre-Yves Oudeyer and his team are studying autotelic curiosity, that marvelous ability to self-generate and self-select our own learning objectives, goals or games.

They design robots to learn like babies…

 If you compare, for example, the most powerful AI systems that most people use today — like those generative AI systems, like those chatbots — they do very impressive things, but they’ve been trained extensively and passively… they are fed datasets by engineers. It’s not at all the way children learn.

We talk about the origins of his curiosity about curiosity, that the brain prefers to explore ideas where it is making progress (learning progress theory), how we study exploration in the lab, language as a tool of exploration, teaching robots to generate their own learning goals, flexibility — and solving problems by forgetting about the problem and just being curious.

Listen to Ep. 338: Make It Your Own: Autotelic Curiosity, with Pierre-Yves Oudeyer

Pierre-Yves Oudeyer, PhD, is the Research Director at Inria and head of the Flowers AI & CogSci Lab in Bordeaux, France. This is his terrific and quite readable review of the curiosity research-to-date and context for his work.

Check out some of the research from Pierre-Yves and his colleagues, start with this nice Scientific American piece and then dig into something like this: Aces: Generating a diversity of challenging programming puzzles with autotelic generative models or MAGELLAN: Metacognitive predictions of learning progress guide autotelic LLM agents in large goal spaces.

Want more curiosity/learning research? Check out these C2BC Classics: Curiosity, Kids & Learning to Learn, with Elizabeth Bonawitz; Curiosity Becomes the Engine for Learning, with Matthias Gruber; Neuroscience Is Taking Us to New Frontiers, with Jacqueline Gottlieb; and Curiosity & Creating Sense in a World of Nonsense, with Saul Perlmutter.

Theme music by Sean Balick; “Cradle Rock” by Nursery, via Blue Dot Sessions.

You can subscribe to Choose to Be Curious on Apple Podcasts and Spotify, or wherever you get your podcasts.