Hacker News Reader

Considering they are a data company with heavy AI departments - not be able to differentiate basic russian bots from users is very embarrassing.

Most problems Google (& DeepMind) have seen success in are not arms races; you're simply trying to accomplish a difficult task.

Dealing with bots, and content moderation in general, is an arms race. You're fighting another person who could be just as smart as you in developing evasion strategies as you are in countering them. Getting it right all of the time is impossible.

adversarial networks - isn't this exactly the branch of AI which is supposed to solve this?

I'm not sure exactly what you mean by 'adversarial networks', it could be GANs or adversarial training.

Either way, yes, there are techniques for improving performance in the presence of an adversary but it doesn't negate the problem; anything you do will be countered eventually. Clever techniques only buy you more time before someone figures out a way around it.

Yes, not embarrassing for them -- but is more clarification for us.

Embarrassing is not the same as surprising