From: Eugen Leitl (eugen@leitl.org)
Date: Sun Apr 07 2002 - 06:02:43 MDT
On Sun, 7 Apr 2002, Sean Williams wrote:
> As an observer to the list, I'm getting a weird sense of double vision
> while trying to assimilate the two current main topics of
> conversation: (1) the difficulty of simulating the behaviour of a
I try to avoid double vision by focusing upon one topic at a time.
> single neuron and (2) the cause and effects of conflicts present and
> past. It seems to me that if even extraordinarily powerful computers
> are having difficulty accomplishing the former, how can we expect any
> genuine accuracy to creep into our personal models of the behaviour of
> immensely large numbers of neurons far removed from us (i.e.
> politicians past and present, from our cultures and others)? I know
Strange notions you've got there.
What have computers do with it? Do you use mathematical proofs when
crossing a street, while avoiding traffic and smiling absentmindedly at a
little kid on a bike? You're still alive, so I presume you're not. Why
should you starting with it, then?
Reality. You're soaking in it. Whether you want it, or not, you're forced
to do a number of realtime decisions in face of faulty data and faulty
models. Something more precise but too late is wrong. There's no helping
to it, unless you don't want to play anymore.
You tacitly assume a complex system will generate even more complex
phenomena at a higher level (trying to avoid using emergence and not
succeeding). If this was always true, science was impossible. Water is
complex, but it has a single freezing point. Superpersonal phenomena are
nontrivial, but unknowable they're not, orelse you'd be forced to do the
equivalent of a dice roll at each decision. Since you don't, they've
clearly got a structure different from random noise.
> that the brain is a powerful neural net designed, in part, to
The brain is a muscle. A substance very like porridge. An organ to cool
your blood. Cavity sealant. Ballast. Fnord repellant. A mysterious
substance to keep you distracted from the truly important issues in life.
> recognise emergent properties in other such nets (i.e. the behaviour
> of people around us), but I think it's worth bearing in mind that no
> simulation we can run is likely to tease out the motives or
> aspirations of anyone in the Palestine/Israeli conflict with any great
> accuracy. It's all guesswork, and getting incensed because someone
Nonlinear systems diverge exponentially. Even in absence of intrinsical
noise (you'd wish) and perfect knowledge (you'd wish) the cost function
raises very steeply. Translation: precise prediction of nonlinear system
trajectories is impossible. (Disclaimer: assuming you're not God).
> else's guesswork disagrees with yours seems no different (to me) to
> two neuron modellers bickering over the outputs of their simulations
> -- especially when neither model may in fact bear much resemblance to
> what occurs in a real neuron itself.
The neuron modellers (as any decent modellers) know better than this. If
you don't validate, you could as well climb back into your virtual navel,
and make home there.
This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:13:18 MST