From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Sat May 24 2003 - 16:26:59 MDT
On Sat, 24 May 2003, Eliezer S. Yudkowsky wrote:
> Let it be noted for the record that, when at Spike's house for the
> post-Gathering Gathering, I criticized Bradbury's MBrain plans on the
> grounds that they "allowed the sun to stay on".
Eliezer, since I don't recall this conversation, I must assume that it
took place after I had left Spike's for Seattle -- which is fine
though it is unfortunate since we didn't get a chance to interact head to
head on this.
My general comment on "allowing the sun to stay on" is that I completely
agree -- it is a stupid waste of resources and suns should be turned off.
The problem is that it requires a huge amount of energy and a relatively
long time (millions of years at least) to turn it "off" (with some
possible exceptions that I cast shadows on in a reply to Spikes message
but I'm not willing to discuss in public until I can convince myself that
the numbers really work).
So *yes* I strongly agree that the sun (or any star powering an advanced
civilization) should be turned off and the energy resources diverted to
a consumption process that maximizes energy production and utilization
efficiency. (Now the best ways to do this involves some very complex
discussions.) I believe I do have some mention in some of the MBrain
papers regarding the need to adapt a system architecture to precisely
*what* the system wants to "think" about.) I believe that architectures
would vary from a dismantled star (consuming its energy in fusion reactors
very slowly), to a common solar system (using a stellar gravity well
to drive fusion processes), to an "existentialist" system where through
some means the energy production/consumption process is so accelerated
that neither the star nor the system (perhaps an MBrain around the star)
could survive. The last alternative might be called destructive
computing (since the information it might generate would involve its
own self-destruction).
> Even if you catch 100% of the Sun's power output,
> the Sun itself is generating many bits of entropy that are not being used
> to write reversible computation outputs to memory, since the Sun is not a
> computing device.
Yes of course. And while I would clearly like to see the resources
go into reversible computational devices and perhaps the greater
production of extropic vs. entropic information -- I *do* not know
what the net-present-value of extropic vs. entropic information is.
Classical economic theories would suggest that a penny of extropic
information now may be worth much more than a pound of entropic
information later.
In case this isn't clear -- pls use the "destructive computing"
situation I have outlined above and detail "approximately"
how many star systems you would be willing to sacrifice to
generate a stable/productive/friendly AI within this star system?
(The point of the question is to generate an evaluation of relative
levels of extropic and entropic activities in the universe and ways
they might be influenced and is not directed at positions of EY or
the SI in any way. [I agree with many, though not all, positions
of EY and the SI as I currently understand them.])
So my questions are more of an inquisitive nature and not a
disputive nature.
> We must thus conclude that a sun plus an MBrain is massively inefficient
> as a computer,
Absolutely -- the question you have not resolved is whether an
MBrain in a million or a billion years can define a computer
architecture using "physics which works" which is significantly
faster/better/(whatever criteria you want to apply that we don't
currently know how to measure) such that 10 minutes of thinking
in a 2nd, or 3rd, or 4th Gen MBrain is equal to 1 million years
of thinking in a 1st Gen MBrain. If that happens to be the case
then one should be burning up stars left and right to get to the
4th Gen Mbrain -- because it is going to really really really
kick butt.
> since the total physical process generates many more bits of
> entropy than it performs useful irreversible computations.
Yep, you have little disagreement from me on this point. Generally
speaking we should *not* be using stars as primary energy sources
(though I don't think this is really proven yet but I'm willing
to accept it as a working hypothesis) and we *should* be using
reversible computations.
The question I would ask is how one might know that the
total (extropic) information content of any web arena is
greater this saturday than next saturday. If one cannot
know that then everthing from allocating ones time to
pricing tickets seems up for grabs.
Robert
This archive was generated by hypermail 2.1.5 : Sat May 24 2003 - 16:37:14 MDT