[ExI] Fwd: Re: AI risks

[Anders Sandberg]

[Delayed post]

On 2015-09-02 09:42, BillK wrote:
> <http://dilbert.com/strip/2015-09-02>
>
> It's funny, but the point about all the effort going into making sure
> that AI doesn't harm humans is --
> How do you control humans that deliberately want to harm humans?

We actually have a fairly extensive infrastructure for that. Think about
it: parents socializing children, reputations, economic incentives to be
pro-social, policing, rule of law, armies and intelligence agencies. By
no means perfect, but actually quite pervasive (consider the part of GDP
going into all that) and effective (see Pinker's "Better Angels" data).

The concern should be: are AI-empowered people a threat that require new
tools or approaches? I think that is true and should be investigated
more deeply (I hope to do some stuff on this myself at our FLI-funded AI
strategy centre).

-- 
Dr Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University