The point at which we have to worry is not "when we start building robots".
It's when robots, from nothing more than raw material, start building themselves.
At that point, they are pretty much immortal (that doesn't mean invincible, however).
Mining operations can be largely automated, transport of raw material to a factory, refinement, fabrication and assembly. But at the moment it's all "instructed", driven by us. When we start to pop out of that loop, then we're in trouble.
But, hell, AI is 50-100 years away. Sure, we can do fancy things blurring pixels together and tell you if there's a banana in an image but - coincidentally - so can purely mathematical statistical models with no "learning" whatsoever. It's almost like that's exactly what they're selling you instead of "intelligence"... odd that...
We're approaching a "sufficiently advanced technology" area where it looks like AI/magic but it's not. Turing test contestants are still naff (you can often find them online if you want to have a play), and that's still in the same area, i.e. not actually intelligence.
Given the acres of supercomputers in the world belonging to some of the richest corporations in the world, with almost limitless talent, capability and resources... Siri, Alexa and Google still do a sucky-job.
Biting the hand that feeds IT
