News for hackers. Stuff that matters.

A comment on Google self-driving car pioneer wants to teach people how to face the future. Parent comment

Re:Google self-driving car pioneer wants to teach people how... by JoshTriplett on Friday September 04, 2015 @01:09PM (#10171663)
> However, this whole "people need more skills" and "we need more jobs" meme feels like a losing fight with reality.

In the long-term, sure: if we can successfully build an AI smarter than humans, it's unlikely any job will survive. But at the same time, no job will need to; there's a balance there. Planning for a post-scarcity economy makes sense, but the plans that work in such an economy don't necessarily work when scarcity still exists.

[ Parent | on HN ]
Re:Google self-driving car pioneer wants to teach people how... by StavrosK on Friday September 04, 2015 @01:11PM (#10171678)
> But at the same time, no job will need to

Why? Because this AI, which is basically just people smarter than us and getting smarter by the day, will decide that its only purpose in life is to serve its idiot masters?

[ Parent | on HN ]
Re:Google self-driving car pioneer wants to teach people how... by JoshTriplett on Friday September 04, 2015 @01:17PM (#10171718)
Because if we build it correctly, it won't be a person at all, and won't have a value function orthogonal to our own. Building something that can self-improve and solve problems doesn't mean we need to build something that has independent will, motivation, or values. And it's a lot safer not to.
[ Parent | on HN ]
Re:Google self-driving car pioneer wants to teach people how... by StavrosK on Friday September 04, 2015 @01:30PM (#10171780)
Sure, but "an AI that's smarter than humans" implies that. I'm sure we'll build many algorithms that are better than us at specific tasks, and we already have.
[ Parent | on HN ]
Re:Google self-driving car pioneer wants to teach people how... by JoshTriplett on Friday September 04, 2015 @02:35PM (#10172115)
Even building algorithms that are better than us at any general-purpose task still doesn't require giving those algorithms agency or independent values of their own.
[ Parent | on HN ]