[extropy-chat] Desirability of Singularity (was Are ancestor simulations immoral?)
mstriz at gmail.com
Tue Jun 6 17:40:16 UTC 2006
On 6/6/06, Eugen Leitl <eugen at leitl.org> wrote:
> On Mon, Jun 05, 2006 at 06:08:42PM -0400, Martin Striz wrote:
> > Gotcha, so you're assuming a computational substrate that has zero
> > energy demands other than for information erasure according to the
> > Brillouin inequality.
> You don't have to erase bits. That's what reversible computing is all about.
> > Understood now, but what's the point? To say that evolution is X%
> > inefficient with respect to some idealized (and impossible) substrate
> > is rather uninformative.
> Why on earth "impossible"? All of technology used to be impossible,
> quite a short while back.
It's a Gibb's free energy thing. You could design a computational
system with a complex set of internal transition states that maps an
input to an output. You could rig it to be spontaneous in the
forward direction when an input is present. But if you want to use it
more than once, you'll have to input energy to reset it.
I suppose you could have a reversible system, but then the probability
of getting the computation done is lower.
> > Are you trying to use that as a canonical metric for comparing any
> > future computational substrate? A perfect computational substrate,
> > i.e. one that also has zero information erasure, runs on zero energy.
> > :) Have I said anything interesting?
> No, you can come quite close to zero energy. In practice, perfectly
> reversible means slow, and perfect doesn't exist.
Very well, I just find it to be a nontraditional use of the term
"efficient." To me that is a measure of the percentage of input
energy that gets used to do the work of the system, versus the
percentage that is lost as heat, etc. By that measure, neurons are
pretty efficient. There's very little heat loss. Your head isn't
warm due to heat loss. It's kept warm on purpose because enzyme
kinetics are optimized for 37 C.
Neurons are complex. However, sometimes complexity can be used to
create efficiency. For example, cellular respiration is a controlled
combustion reaction, whereby the reduction potential, rather than
being lost as heat, is captured as energy stepwise through many
intermediates. Efficiency is maintained.
> > I thought a useful calculation would be to determine how efficient you
> > could make cells for computation, but that of course is unknown.
> > Neurons, as it happens, are probably greater than 90% efficient with
> > respect to /their design/, i.e. there's little heat loss.
> Neurons just happens to be cells with a pretty high metabolic rate.
> Calling them optimal in regards to computation is a pretty weak joke.
Not ops/unit energy, but see above.
More information about the extropy-chat