BROADCASTS.com

b'

Imagine, for example, an AI that\\u2019s trained to identify cows in images. Ideally, we\\u2019d want it to learn to detect cows based on their shape and colour. But what if the cow pictures we put in the training dataset always show cows standing on grass?

In that case, we have a spurious correlation between grass and cows, and if we\\u2019re not careful, our AI might learn to become a grass detector rather than a cow detector. Even worse, we could only realize that\\u2019s happened once we\\u2019ve deployed it in the real world and it runs into a cow that isn\\u2019t standing on grass for the first time.

So how do you build AI systems that can learn robust, general concepts that remain valid outside the context of their training data?

That\\u2019s the problem of out-of-distribution generalization, and it\\u2019s a central part of the research agenda of Irina Rish, a core member of the Mila\\u2014 Quebec AI Research institute, and the Canadian Excellence Research Chair in Autonomous AI. Irina\\u2019s research explores many different strategies that aim to overcome the out-of-distribution problem, from empirical AI scaling efforts to more theoretical work, and she joined me to talk about just that on this episode of the podcast.

***

Intro music:

- Artist: Ron Gelinas

- Track Title: Daybreak Chill Blend (original mix)

- Link to Track: https://youtu.be/d8Y2sKIgFWc

***

Chapters:

\\n
• 2:00 Research, safety, and generalization
\\n
• 8:20 Invariant risk minimization
\\n
• 15:00 Importance of scaling
\\n
• 21:35 Role of language
\\n
• 27:40 AGI and scaling
\\n
• 32:30 GPT versus ResNet 50
\\n
• 37:00 Potential revolutions in architecture
\\n
• 42:30 Inductive bias aspect
\\n
• 46:00 New risks
\\n
• 49:30 Wrap-up
\\n