Podcast 317: Chatting with Google’s DeepMind about the future of AI
Teaching AI to master games without knowing the rules may help to lay the foundation for more general intelligence in real world environments.
Ben Popper
Director of Content
This week we chat with Julian Schrittwieser, a staff software engineer at DeepMind, the AI lab acquired by Google in 2014. He is one of the authors of a recent paper on MuZero, an AI program that mastered “Go, chess, shogi and Atari without needing to be told the rules, thanks to its ability to plan winning strategies in unknown environments.”
You can find the paper on MuZero here.
He blogs at Furidamu and can be found on Twitter here.
The story on drug discovery powered by AI can be found here.
Tags: artificial intelligence, deep learning, deepmind, the stack overflow podcast
4 Comments
Hard to talk about Google’s AI deficiencies and flaws not mentioning Margaret Mitchell and Timnit Gebru.
When did you realise racism is vital for the success of Google?
How many more exciting firings are you planning in 2021?
What brilliant career success a man like Julian Schrittwieser can achieve by avoiding being black and woman?
think this is the first time ive heard cassidy on the podcast. she was super funny. ++ for her!
Sounds like the real problem with generalized application of AI will be the development of meaningful metrics for success. For example, we asked this AI to design a model for education, and its success is evaluated against a set of weighted and prioritized values. Who chose these metrics and how? What agendas drove them? What ignorance or prejudices were in play? Were they even the appropriate data types — scalar values, booleans, etc.?
It’s less a question of AI making people useless than it is of people making AI useless.
In the transcript says the deep learning library with a numpy API he uses is called “Drax”, but actually is “Trax” (https://github.com/google/trax), for anyone interested.