[extropy-chat] limits of computer feeling
eugen at leitl.org
Tue Mar 20 20:51:01 UTC 2007
On Tue, Mar 20, 2007 at 04:20:26PM -0400, Rafal Smigrodzki wrote:
> ### Evolution is computationally highly inefficient. It takes
You still have to guess what the other guys will come up at
the next iteration. If you don't guess quickly enough or
accurately enough, you've lost. Regardless of which mode
of evolution you utilize.
And right now the embodied computation packs a lot of punch,
especially if scaled up to the size of an ecology.
> mountains of dead bodies to come up with even trivial improvements. A
> search through the design space of an enzyme may take millions of
> years and staggering amounts of energy. Progress occurs as a
> side-effect of humongous waste.
Yes, but the process doesn't attempt to optimize a particular enzyme.
It's a side effect.
> Intelligent design of offspring means using highly efficient
> computational algorithms running on specifically designed hardware.
Notice: there is no such specifically designed hardware yet.
And you're computing against a Darwin in machine-phase, not today's
> Searches through design spaces of enzymes may be soon feasible using
> only kilowatts of energy. It will be possible to design whole new,
Right now physical modelling takes a lot of cycles. And these are
extremely inefficient cycles.
> optimized metabolic networks from scratch, rather than cobble together
> old, only marginally appropriate ones.
> It is true that the designers do not have a god's eye view of the
> fitness landscape but then evolution doesn't either. In fact, where
> the designers can look ahead to the next few kinks in the metabolism
> or synaptic wiring, evolution is perfectly blind. There is absolutely
> no foresight in this process, mistakes are repeated millions of times,
The difference between Darwin and Lamarck is largely that one blind
trial is embodied, and the one is virtual, and only the winner is
embodied. Latter machinery has a cost which brute-force faulty replicators
don't have to bear.
No, I don't think Lamarck will will over Darwin at any time.
> mutational moves are made with the same likelihood independently of
> their results on fitness.
> Lamarckian agents will be ultimately judged by their fitness, I agree,
> but given their improved computational efficiency, they should handily
I can do both Lamarck and Darwin at the same time. There is a
continuum between them -- assuming that I can test things virtually,
and don't have to test them all embodied.
> outcompete Darwinians, as long as this is what they desire.
Anything which doesn't think competitiviness is important on the long
run isn't going to matter, on the long run. Space is big, you have
to want it to colonize it.
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
ICBM: 48.07100, 11.36820 http://www.ativel.com
8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
More information about the extropy-chat