Could an AI win a Nobel prize? That’s the provocative question a team of scientists are pondering after putting GPU-powered deep learning to work to accelerate their science, with impressive results.
“AI is beating conventional methods,” Paul Wigley, a doctoral student at the Autralian National University told host Michael Copeland on this week’s edition of our AI Podcast. “The more we apply it to physics, the more it will advance,” Wigley said.
Curious about the scope of deep learning in physics-related experimental applications, Wigley was a member of a team that saw an opportunity to use AI to produce a Bose-Einstein condensate — a state of matter in which atoms are cooled to temperatures close to absolute zero and a starting point for other measurements.
Retained in massive metal chambers, atoms are essential stored in cup-like structures with walls during an evaporating cooling stage. Traditionally, the walls are lowered by human scientists through lasers to allow hotter atoms to escape. Wigley explained his team decided to use AI to adjust the power of laser beams. The results were surprising.
Normally the laser’s output would decrease not increase, but the AI found a way to modulating the laser beams to create the condensate, Wigley explained. “It begged the question: Was there more physics that we didn’t necessarily know about that the AI could find or could it make discoveries about science?”
AI Podcast: Hold the Mayo – ‘Not Hotdog’ Brings App from HBO’s Silicon Valley to Life
Hungry for more podcasts? If you missed last week’s episode, listen to Tim Anglade share his experience developing “Not Hotdog,” an app that appeared as a joke on HBO’s Silicon Valley TV show and is now available on iOS and Android stores.
How to Tune in to the AI Podcast
The AI Podcast is available through iTunes, DoggCatcher, Google Play Music, Overcast, PlayerFM, Podbay, Pocket Casts, PodCruncher, PodKicker, Stitcher and Soundcloud. If your favorite isn’t listed here, email us at email@example.com.
Featured image credit: Adam Baker, via Flickr