What I think is wrong with Eli's current approach

From: Marc Geddes (marc_geddes@yahoo.co.nz)
Date: Sun Oct 24 2004 - 23:30:54 MDT


Eli said>

>Someone other than me needs to talk to Marc Geddes
before he snaps completely and becomes another
Mentifex, complete with incomprehensible ASCII
diagrams. I don't have the time and I don't have the
tact, but Geddes was once a promising mind and there
might be some way to pull him out of this.

Heh. The philosophical schematic was designed only to
make sense to someone who already has an inkling of my
general theory. It wouldn’t mean much to you, but it
does to me. Be assured I do have a general theory
that makes some degree of sense. I suggest you come
back and take a another look at my schematic after
you’ve been working on FAI theory for 10 more years ;)

But O.K Eli I promise that I will not continue to
discuss my own ill-formed ideas on Sl4 (excepting this
one post). I am simply going to try to explain, as
best I can, what I think is wrong with your current
approach.

My schematic on the SL4 wiki:

http://www.sl4.org/bin/wiki.pl?FundamentalTheoremofMorality

You'll notice in my schematic the entry for 'POLITICS'
and 'POSSIBILITY' in my matrix reads 'MARKET', which
in my explanation of terms I said was referring to CV
(Collective Volition is a kind of highly sophisticated
'futures market'). In fact you'll see that a lot of
the issues that Eli is dealing with I have placed in
the POLITICS row of the Cognition matrix, instead of
the ETHICS row.

In short I think that CV (Collective Volition) is 'a
nice place to live', or *a good political system*. It
is *not*, I believe, something that can be
comprehended by or embodied in a single agent. So CV
is the operational *global outcome* of morality, *not*
something that a singleton AI *does*. What exactly
do I mean? Well, I'm really saying that I think that
the whole top-down approach *can't work*

Eli's requirement that his AI *not* be sentient should
be the tip-off that there is something highly suspect
and peculiar about his proposed RPOP. Should actual
consciousness emerge in simulations of sentients, then
RPOP is immediately stymied, since it would not be
able to make effective simulations of sentients
without violations of *person-hood*. Worse, a
conscious (sentient) RPOP would immediately run into a
problem with self-reference. A conscious RPOP would
be a *person* itself, and a suitably generalized
definition of *person-hood* would end up with the RPOP
having to include its own volition in the calculation
of CV. This leads a fatal infinite regress.

Is general intelligence without sentience possible? I
say not, not in any *practical* sense. Of course we
can imagine a theoretical general intelligence with
infinite computational power. In that case, I agree
that general intelligence without sentience would be
possible. One would simply take a pure Bayesian
reasoning machine, capable of duplicating any kind of
intelligence without sentience by burning up as much
computational resources as it needed.

But in the real world, there can be no such thing as
*infinite computational resources*. Any real world AI
will only have access to *finite* computational
resources at any given time. General intelligence
would require *useful computational short-cuts* in
order to do useful things in real-time. Theoretically
ideal Bayesian reasoning won't work, because it will
quickly run into computational intractability for
complex problems. So all finite resource AI's would
need *specialized computational short-cuts*. And
these *computational short-cuts* I maintain, are what
*necessarily* give rise to qualia (consciousness and
sentience).

To sum up: Any RPOP would quickly run into
computational intractability if it stuck with pure
Bayesian reasoning. It would be forced to resort to
*computational short-cuts*. These would, I claim,
inevitably give rise to consciousness. With qualia
present the RPOP would now be a 'Person'. A suitably
generalized definition of 'Person-Hood' would result
in the RPOP being forced to include its own volition
in the CV calculation. This would give rise to an
infinite regress. Ergo, Collective Volition cannot be
calculated by a singleton RPOP and the entire top-down
approach is flawed.

I should make it clear that I *do* agree that CV
(Collective Volition) is the ideal *political system*.
 That is, I agree that CV is *a nice place to live*.
But I disagree that CV is something capable of being
embodied in a singleton RPOP and imposed from the
top-down. Eli's mistake is his insistence on the
top-down approach. He has mistaken a *distributed
system* (Collective Volition) for a mind. But in fact
CV is not a singleton.

Under my theory, all working FAI's are necessarily
sentients which assign themselves *Person-hood*
status. No singleton FAI can possibly implement
Collective Volition (since any FAI is itself
*included* in what constitutes Collective Volition).
None the less, CV would still represent an ideal
*political system* for sentients, which the FAI's
would try to act in harmony with.

Under my theory no singeton FAI can fully calculate
CV, but it *can* still obtain some degree of
understanding and determine which actions are and are
not in harmony in CV. That is, FAI could still
perform calculations about CV sufficient to establish
a sort of 'futures market' to help determine which
actions were *Friendly* and which were *Unfriendly*.

CV places constraints on permissible sentient actions.
 But it's a distributed *global system* and *not*
something that can be embodied in a Singleton as
Eliezer thinks.

Of course I've banging away with my objections on SL4
for a couple of years now ;) Recall that I've always
said that;

(1) (Practical) general intelligence without
sentience is impossible
(2) Completely selfless AI is impossible

Now though, I think my objections are stronger because
I've got some plausible reasons for them and my own
general theory of FAI.

Eliezer. And me. One of us has to be right out of
this, the other has to be wrong (Marc chuckles to
himself and nods his head). The time is fast
approaching when we'll find out who's who...

=====
"Live Free or Die, Death is not the Worst of Evils."
                                                    - Gen. John Stark

"The Universe...or nothing!"
                            -H.G.Wells

Please visit my web-sites.

Sci-Fi and Fantasy : http://www.prometheuscrack.com
Mathematics, Mind and Matter : http://www.riemannai.org

Find local movie times and trailers on Yahoo! Movies.
http://au.movies.yahoo.com



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:49 MDT