Kellin Pelrine - How He Crushed A Superhuman Go-Playing AI 14 Games To 1 #34
Description
Our guest today is Kellin Pelrine, Research Scientist at FAR AI and Doctoral Researcher at the Quebec Artificial Intelligence Institute (MILA).
In our conversation, Kellin first explains how he defeated a superhuman Go-playing AI engine named KataGo 14 games to 1. We talk about KataGo’s weaknesses and discuss how Kellin managed to identify them using Reinforcement Learning.
In the second part of the episode, we dive into Kellin’s research on building practical AI systems. We dig into his work on misinformation detection and political polarisation and discuss why building stronger models isn’t always enough to get real world impact.
If you enjoyed the episode, please leave a 5 star review and subscribe to the AI Stories Youtube channel.
Follow Kellin on LinkedIn: https://www.linkedin.com/in/kellin-pelrine/
Follow Neil on LinkedIn: https://www.linkedin.com/in/leiserneil/
————
(00:00 ) - Intro
(01:54 ) - How Kellin got into the field
(03:23 ) - The game of Go
(06:10 ) - Lee Sedol vs AlphaGo
(11:42 ) - How Kellin defeated KataGo 14 -1
(26:24 ) - Using AI to detect KataGo’s weaknesses
(37:07 ) - Kellin’s research on building practical AI systems
(43:10 ) - Misinformation detection
(49:22 ) - Political polarisation
(54:39 ) - ML in Academia vs in Industry
(1:06:03 ) - Career Advice
United States
