vogon
@vogon

good news! a human beat AlphaGo1.

bad news! he's an amateur go player2 who works for an AI research think tank, who used a technique that was discovered by another machine learning system which discovered the technique after more than a million games of go.

unfortunately I worry that the future holds a lot of stories like this: a machine learning system deployed on the basis of some thin claim to better-than-human performance, with blatant flaws that lie dormant until some other group of ML researchers have a monetary incentive to prove otherwise.


  1. technically he beat a publicly available go player which is considered to be at par with AlphaGo, not AlphaGo itself, because Google doesn't allow people to have access to AlphaGo.

  2. the article says he's in the second-highest amateur rank; I assume they mean 2-kyu.


amydentata
@amydentata

I find this heartening, personally. At this point I feel like machine learning is going through a cryptocurrency-esque train of hype that will be soon followed by crash after crash. Unlike cryptocurrency, ML will still retain some neat applications when the dust settles, but it won’t be the Total Revolution a lot of people are forecasting. The best applications will be niche and/or narrowly specialized (3D animation blending, video compression, protein folding), and after the novelty wears off, AI media generation will settle into something akin to a new feature in Photoshop.


You must log in to comment.

in reply to @vogon's post: