Fabrizio Musacchio

🧠👀 Merits of Curiosity. Gruaz et al. (2025) compare 6 intrinsic motivation signals in simulated agents. Turns out: curiosity isn’t one thing. The best-performing agents combine #novelty and #InformationGain, revealing how environment and goal shape exploration. Cool #modeling paper with implications for #AI, #RL, and theories of human #curiosity.

🌍 https://direct.mit.edu/opmi/article/doi/10.1162/opmi.a.9/132758

#NeuroAI #CompNeuro

Aug 29 at 7:01pmWeb
Show threadandrecattaAug 29
@pixeltracker but curiosity – in animals– is not just reward-seeking.
Plain transfer of ill-defined biological mechanics has already led to many tech architectures that didn’t delivered so much.
So, my question is, how to learn from nature avoiding reductions?
Just thoughts :)
Show threadFabrizio MusacchioAug 29
@andrecatta Great point! You’re right. Biological curiosity is multifaceted and not merely reward-seeking. I think that’s exactly why I find the paper’s finding interesting: combining novelty and (!) information gain worked best. Maybe it’s a step toward more nuanced models without reducing nature to a single drive.