On 29/10/2014 14:38, Matt Mahoney via AGI wrote:
On Wed, Oct 29, 2014 at 5:50 AM, Tim Tyler via AGI <[email protected]> wrote:
Really? You think machine progress will slow down once humans have been
replaced be their artificial offspring?
Yes, but we are not there yet. The biosphere computes 10^33 DNA, RNA,
and amino acid operations per second on 10^37 bits of DNA. Global
computing capacity is currently about 10^20 OPS on 10^22 bits. It will
take about 75 years to catch up at the current rate of Moore's Law
(doubling every 1.5 years). After that, progress will slow down
because of the limits of physics.
Well, sure - but that's a bit different from: "The apparent rapid progress in
machine
evolution is due to the transfer of human knowledge to machines (design, coding,
training)."
No doubt existing knowledge contributes to progress - but machines will
probably be able
to make good progress in the future without much in the way of humans. It's
hard
to say how much existing human knowledge is helping to spur the rate of
progress.
Having humans in the loop may well slow things down more than it speeds them up.
One bit per generation?!? I thought we already debunked that figure.
There is no such limit.
One bit per generation with respect to knowing what improves
reproductive fitness. I realize that information can be transmitted to
offspring epigenetically and through language. That's not the same
thing.
A thought experiment should disprove that. Imagine a musical
supervisor allows organisms to reproduce if they play back the
correct Simon sequence [http://en.wikipedia.org/wiki/Simon_%28game%29] -
and they add an extra 5 notes to the correct Simon sequence in each
generation, tell the parents, and allow them to transmit the knowledge
to their offspring (either culturally or via genetic engineering). That's ten
bits per generation. Want more bits? Just add more notes.
This surely qualifies as "knowledge that improves reproductive fitness"
in the specified environment. There isn't much of a limit to how fast
such knowledge can grow.
You might think that advanced technology can speed up testing. But
once you start thinking in terms of computation rather than magical
intelligence, you'll realize otherwise. Simulating even simple
chemistry requires solving Schrodinger's equation, which is
exponential in the number of particles unless it is run on a quantum
computer. This is why there are no programs that can input a chemical
formula like "H2O" and output the properties of that chemical such as
its melting point. I expect that we will develop better quantum
computers, but remember that what you do to program a quantum computer
is to build a physical system that obeys the same equations. It's not
going to run significantly faster than doing a physical experiment
like putting a thermometer in ice water.
That's sounds pretty pessimistic - but it also sounds wrong. Simulation has
already
sped up all kinds of thing. We simulate bridges, skyscrapers and areoplanes
before
we build them. Virtualization really *does* speed up testing. You don't need to
simulate at the level of the Schrodinger equation for simulation to be useful in
speeding up development. Rapid prototyping is a real phenomenon.
--
__________
|im |yler http://timtyler.org/ [email protected] Remove lock to reply.
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
