From: Eugen Leitl (eugen@leitl.org)
Date: Tue Dec 02 2003 - 04:38:03 MST
On Mon, Dec 01, 2003 at 05:28:59AM -0800, Lawrence Foard wrote:
> > It's a computer, Jim, but not as we know it.
>
> I assume your talking about neural nets, not magical mystical macro
> quantum computers?
No, I mean the real thing. A neural net (there are some 100 types
of various ANNs) is an abstraction, capturing
only few relevant aspects of the actual tissue. It is a good reality check
to see how far ANNs are remote from something they're supposed to
be a decent model of.
In biology, the system and the representation have coevolved. There is no
separation between state and the physical layer, they're amalgamated.
The system uses noise and dirt effects to do computation. It turns
ugly warts into actual advantages.
> Yes. Thats the problem, but at all levels. In the cell the context becomes
> very important, first of all the DNA decoding machinery which luckily for
> us is fairly straight forward in its action, then after that the physics
> of proteins, as well as the cell environment in which they operate. And
> thats just one cell not a complete human.
>
> So still most of biology must be empirical.
We can't even image a whole cell at molecular resolution yet!
Not even small representative patches of it. It is really quite
pathetic. We don't even have the data to put into our computers,
and nevermind our computers are far too feeble to probe any
interesting scales on which biological processes occur.
We can't even do protein folding yet! Utterly hilarious.
> True. But even with those tools many answers will be found empirically
> through simulation, rather than direct mathmatical derivations from first
There are people who would burn you at the stake for mouthing 'empirical'
right next to 'simulation' :)
> principles. The world already has a large heritage of such mental
Mathematical derivations from first principles are something you do
physics with. They're really quite irrelevant to biology (you can
derive forcefield parameters from QM, but the quality is so lousy
you'd be better advised to use empirical measurements to get at it).
Nevermind that simulations are next to useless in biology.
It's more a crude mockup of the real thing; it has only relatively
recently started generating new knowledge -- eyed suspiciously,
because you frequently can't validate the observation in machina
empirically.
> observations, regrettably they are often overlooked because they have been
> given the label 'spirituality', rather than 'psychology'.
I can't parse either; both are irrelevant to current science.
> I don't believe in a human like God, its an absurd notion. Yet I also know
> that belief in a God works for people. I suspect why it works as unPC as
> it sounds is because God is a surrogate imaginary alpha-male. An
> alpha-male so powerful that his authority is never open to challenge,
> gaining a permanent state of submission and surrender from true believers.
> But also a dangerous thing, because anyone claiming to represent God can
> have unquestioning followers.
>
> The regrettable thing about modern psychology is that it started from
> scratch. It didn't look at what actually works and adopt that for the
> scientific age.
If you want to build AI, you can completely throw out psychology
(or philosophy, for that matter; a remarkably sterile are of human
inquiry) out of the window.
You'd do much better with neuroscience, or building IT systems
in simulators from first principles. We don't have nearly enough
performance for that, though, so for time being the best we can do
is to use neuroscience to watch minds at work.
-- Eugen* Leitl leitl
______________________________________________________________
ICBM: 48.07078, 11.61144 http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
http://moleculardevices.org http://nanomachines.net
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT