Re: Why playing it safe is the most dangerous thing

From: sam kayley (thedeepervoid@btinternet.com)
Date: Fri Feb 24 2006 - 19:23:38 MST


----- Original Message -----
From: "Peter de Blanc" <peter.deblanc@verizon.net>
To: <sl4@sl4.org>
Sent: Friday, February 24, 2006 5:15 PM
Subject: Re: Why playing it safe is the most dangerous thing

> On Fri, 2006-02-24 at 06:04 -0500, Ben Goertzel wrote:
> > * If we launch a Singularity before the jerks in power figure out
> > what's up, we have a 50/50 or so chance of a good outcome (by the
> > Principle of Indifference, since what happens after the Singularity is
> > totally opaque to us lesser beings)
>
> But a 50/50 outcome is not the null hypothesis. If the Singularity is
> really totally opaque to us, then we should imagine that all universe
> states are equally probable post-Singularity. The vast majority of
> possible configurations of matter do not contain human life, so the null
> hypothesis is that humans almost certainly cease to exist.
>
> The Singularity is not totally opaque to us because we expect it to be
> the result of goal-oriented cognition, so we can expect the post-
> Singularity world to maximize some utility function. If you're blindly
> rushing into the Singularity and not trying for FAI, then you have very
> little knowledge about that utility function - it may as well be random.
>
> I have a hard time believing that a randomly-selected utility function,
> when maximized, could result in human life, because humanity is very
> complex.
>
A utility function selected randomly with respect to what distribution? A
decision system describing the world only on the level of atoms would be
blindingly unlikely to choose a configuration corresponding to continued
human existence, but would also not be capable of hard takeoff. An AI
capable of recursive self improvement must have a concept of mind, and if it
ends up with a random goal, something that includes the utility of systems
like itself as part of its utility function (not necessarily with the sign
we would want) seems to me although unlikely, more in the range of winning
the lottery than a whirlwind in a junkyard assembling a functioning
airplane.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT