In the latest episode of Approximately Correct, reinforcement learning legend Rich Sutton (Amii Fellow, Canada CIFAR AI Chair & Chief Scientific Advisor) talks about what he thinks is holding AI research back.

Sutton argues that the prevalent approach in AI—particularly the focus on linear static learning models—has made incredible strides but is now limiting further progress. Instead, he says it lacks the capacity for long-term, adaptive learning. “I feel the aesthetics of the field have changed. The field wants to focus on what they can do instead of noticing what they can't do … it's just that simple, you know, we can do certain things, and so we work on those,” he says.

In the episode, Sutton also shares his belief that researchers might soon have a fuller understanding of intelligence and the wide-ranging benefits that understanding could have for society. Finally, he reveals a bit about how he develops his research ideas and how he chooses what to work on.

Approximately Correct: An AI Podcast from Amii is hosted by Alona Fyshe and Scott Lilwall. It is produced by Lynda Vang, with video production by Chris Onciul.

Listen and subscribe to Approximately Correct on Spotify, Apple Podcasts, or your favourite podcast platform.