From: Eliezer S. Yudkowsky (email@example.com)
Date: Wed Feb 06 2002 - 08:10:27 MST
Gordon Worley wrote:
> On Wednesday, February 6, 2002, at 03:17 AM, Eliezer S. Yudkowsky wrote:
> > Mm, that sounds like circular logic to me. The Sysop is what supposedly
> Yes, I should have made it more clear. I am suggesting that the Sysop
> is a tautology,
That is impossible. A Sysop is not a philosophy. A Sysop is a physical
system that may or not exist at some point in our future.
Our concept of the Sysop may be a tautology. If any of our concepts have
the cognitive property of tautology than they must automatically be wrong,
or at the least must automatically possess no force as a rational
argument, with respect to the external referent of the Sysop.
We aren't doing philosophical exploration for the sake of philosophical
exploration, we're taking current knowledge and projecting it forward.
Arguing that something is demonstrated "by definition" in the Sysop
Scenario turns it all into a philosophical game; what matters are simply
those things that are likely to actually appear within our own future.
This is not an attempt to determine what happens if some hypothetical
entity existed with an arbitrarily defined set of defined powers. This is
an attempt to extrapolate events that may actually occur to humanity at
> but this makes sense and isn't really bad since, if the
> Sysop is basically making it impossible to violate someone's volition
> (or whatever), then ve is setting up a system where ve is always
> unhackable. In short, the Sysop sets the rules, so it's very easy to
> make sure that not being hacked is in those rules.
1) "In short, the Sysop sets the rules"
3) "it's very easy to make sure"
4) "that not being hacked is in those rules"
and we have the objections:
1) The degree to which the Sysop sets the rules is precisely the point in
2) The "so" is therefore arguing from the point of the dispute;
3) Whether it is "easy" to prevent high-level hacking given sovereignty
over low-level operations and superintelligence is again precisely the
point of dispute;
4) Whether it is possible to implement the high-level goal of "not being
hacked" in the low-level "rules" is again the dispute.
> But, as I mentioned, just because that's how it works in theory doesn't
> means that attacks are impossible.
It can't work that way "in theory" - who invented the theory? What was
their justification for hypothesizing this as part of a real Sysop
Scenario? At what point did someone, extrapolating forward from current
knowledge, say "The Sysop sets the rules"? If you want to rely on "The
Sysop sets the rules" to argue something, your argument can't be any
stronger than the reasons behind the person saying "The Sysop sets the
rules". The first rule of competent philosophy is to never argue from
definitions. If you want to determine whether the ability to "set the
rules" is strong enough to accomplish something, you look at the reasons
to think that a Sysop "setting the rules" is a possibility in the first
place, and this tells you what you can rationally say about the details
and specifics of these rules and the way they are set.
We have some grounds to think that a superintelligence might be able to
get low-level control over all local material reality, because we can
visualize this as the result of nanotechnological competence by a
singleton SI. When we say "make the rules", what we really mean is
"control reality on a low level". We are now asking the question "Does
the ability to control reality on a low level suffice for immunity to
perversion attacks?" You cannot generalize "the ability to control
low-level reality" into "making the rules", and then argue from this
generalized definition that "prevent perversion attacks" is a subcategory
of "making the rules". *That* is just a philosophical bait-and-switch.
-- -- -- -- --
Eliezer S. Yudkowsky http://intelligence.org/
Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT