Building an AI physicist: Max Tegmark at the 6th FQXi Meeting

Ask not what AI can do for you聽- ask what you can do for AI. That was the challenge that Max Tegmark (cosmologist at MIT and one of FQXi's scientific directors) laid down to his fellow physicists at the recent FQXi meeting in Tuscany.

AI systems have been extremely successful -- but there are still problems. For instance, algorithms used to help make decisions in the justice system have been found to display racial bias -- trained in from biased datasets. These black box systems lack transparency, and thus they are hard to trust. Tegmark asks whether physicists posses skills that could help to train AI to do better -- and crucially make them trustworthy. One of his goals it to create _intelligible artificial intelligence.

This is a very quick placeholder post because I know that our blogger George Musser has been thinking quite deeply about Tegmark's talk and will have some sophisticated views to share soon. But I wanted to alert you that audio from Tegmark's talk has now been posted to the podcast, or at least part of it has.

Tegmark spoke about two papers in his presentation. The edited version of the audio that I have posted only includes the second. The reason is a mundane one -- he referred heavily to a video that you can't see on the podcast when discussing his first paper, "Towards an AI physicist for unsupervised learning," so I chopped that section out. To find out more about that, you'll have to wait for the video.

In the meantime, you can still hear the bulk of his discussion, firstly about the various successes and failures of AI, and what we can do about them. And you'll hear Tegmark discuss in-depth, his work written up in the paper: "AI Feynman: a Physics-Inspired Method For Symbolic Regression." In this project, with his student Silviu-Marian Udrescu, Tegmark wrote a code that used neural networks to "discover" (or to re-discover) 100 equations taken from The Feynman Lectures on Physics.

Enjoy Tegmark's talk -- though my apologies, the sound levels are a little shaky on this one. Slides are available here.