[leplen]

A lot of the stuff described here sort of doesn't make sense. It requires idiot savant AIs that can figure out how to manipulate humans, but that we can only train with simple integer based reinforcement learning techniques.

The introduction of artificial agents is definitely a step increase in the complexity of society, and it will introduce new, and potentially catastrophic failure modes, but I find this story of AIs conspiring to rob a bank to be a fairly unimaginative take. Even if we grant the possibility of conspiracy of AIs, something that looks like insider trading is a lot more likely.

[jopsen]

> something that looks like insider trading is a lot more likely.

Or maybe, our AI overlords would do legitimate hard work, earn money and buy all the companies we work for.

Presumably through a complex web of intermediaries, so that humans never notice.

Maybe, they don't directly control us, but just vote for board members that drive us in the direction they want us to go.

Would make for sci-fi... in practice AIs won't matter anytime soon.