AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
How to Beat a Machine Learning System That Plays Go
As we get better and better machine learning systems i think they're going to remain unnatural and weird. We should approach them with caution because our intuitions about what kind of mistakes are possible don't transfer necessarily from human mistakes to these artificial ones. There was a paper relatively recently looking at ways that you might be able to beat a really good go playing program using a relatively simple strategy. It's already being true for a little bit now that the really best go programs can be the best human players but also this has a cultural cache within the research community.