[ExI] Fwd: Re: AI risks

[Anders Sandberg]

On 2015-09-05 23:25, BillK wrote:
> In the Dilbert cartoon the boss was instructing the robot to kill a
> competitive CEO.
> i.e. How do control humans from deliberately misusing AI?
How do we control them not to deliberately misuse cars, software or 
corporations?

It seems that we have some decent ideas for this, but that we need to 
improve them for technology that is (1) hard to detect the use of, (2) 
gives a significant advantage to the users over the non-users. This is 
similar for other technologies like biotechnology (small labs give 1), 
or powerful atomically precise manufacturing (could give 2).

We might consider making 1 harder by better monitoring, but 2 is tricky 
after the fact: we may need to have better ideas for what capabilities 
give real power and handle them more proactively.


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