AlphaGo Zero is great, but hold on: self-play is one of the oldest ideas in ML, and humans take far less than 5 million games to master Go.
I’ll assume it’s a fair critique about the novelty of the system and claims that AlphaGo Zero is a big step toward artificial general intelligence, but I also think it glosses over the promise of a system like this to solve meaningful problems in other areas.
Mostly, this is a question of scale. AlphaGo Zero played 5 million games and became the world’s best-ever Go player (human or computer) in just 3 days! That’s what’s so powerful here. Maybe in some other field, you take these techniques and run billions of simulations over, say, 3 weeks. The point is that no single human or team of humans could carry out that scale of research in that time frame.
Frankly, I don’t care if AI systems ever reach human-level general intelligence, as long as we can use them to achieve super-human results on certain things that will help humans lead better lives.
And here are three more stories you probably should read today:
HashiCorp raises $40M for its cloud infrastructure automation services (TechCrunch): The plan is to have Hashicorp CEO Dave McJanney CEO on the podcast this week, so hopefully we can dive deeper into the company’s strategy then. I’m very curious to find out if there’s more momentum (and opportunity) around something like Terraform which is very well known but low-level, or something like Nomad that sits up the stack and has to compete for mindshare with Kubernetes.
New version of Microsoft’s Azure Container Service offers managed Kubernetes, with free clusters (GeekWire): Speaking of Kubernetes, Microsoft’s strategy there continues to impress. If VM-based IaaS was Round 1 of the cloud fight, Amazon Web Services won it going away. But if containers/Kubernetes/cloud-native is Round 2, then Microsoft is pulling out all the stops to see that it’s at least competitive – if not the clear winner.
How Pinterest uses AI to learn (and sell) your style (VentureBeat): I cannot point out enough times how spot-on the Pinterest case study is for computer vision and deep learning, in general. It seems like a nearly perfect combination of data, customer intent and algorithms, all leading toward happy users, happy advertisers and a happy Pinterest. Still, I don’t think I’ll ever use it ;-)