From: Eugene Leitl (Eugene.Leitl@lrz.uni-muenchen.de)
Date: Thu Aug 15 1996 - 05:24:15 MDT
On Wed, 14 Aug 1996, Anders Sandberg wrote:
> On Wed, 14 Aug 1996, Eugene Leitl wrote:
>
> [ signs for the slow advent of parallelism ]
>
> The big question is how to use maspar in an useful way. How do we run a
> word processor or matlab in a maspar network?
That was my point exactly: the problem is not hardware. Even von Neumann
investigated introducing parallelism in one of the first machines -- but
he quickly grew disgusted since he kept running into apparently
unsurmountable task organization and coordination problems (a software
problem!), so he abandoned this route very early.
It is interesting to note, that the word 'computer' formerly denoted a
human clerk in charge of computations. In a MIT Press essay I saw a
thesis essentially telling that the advent of digital computers could only
happen so fast because it fell on fertile, already well ploughed ground:
the venerable tradition of organising groups of human computers which
executed sequential algorithms. Astronomical calculation, ballistic
firing tables for artillery, etc.
Parallelism has always been a blind spot in the Western thinking, all the
way since the antique philosophy. Only very rarely an intrinsically
parallel algorithm is being uneasily contemplated, _as if the world was
sequential_, and not vice versa :(( .
How can we teach old programmers new tricks? How?
> This seems to be similar to the problems with process migration and
> distributed computing, although in this case one can (probably) assume
> identical processors at least.
Yes. But at least there is _some_ progress: monolithic mainframes are
quite dead, all current supercomputers are either multi-node machines or
workstations clusters running PVM. Since PCs recapitulate the evolution
of mainframes, we'll see true MIMD machines with non-shared memory within
the next 5-10 years. (At least I hope we will see them. I fully expected
maspar machines to appear by 1987. Ack.)
> I think it is mainly a programming problem - we are not very good at
> creating parallel algorithms, and most obvious ways to do things for us
> are serial when they should be parallel.
Xactly. Just consider Taos
http://www.tao.co.uk/ (though updated, still a bit thin on information)
A revolutionary product if I ever saw one. Enthusiastically reviewed by
Dick Pointain of Byte at least 2 times. Has a 12 kByte nanokernel, is
fully hardware independant since based on a VM (hmm, just where did I saw
this feature recently?), etc. etc. etc. Fantastic product. Ok, their
marketing is lousy, but still no response for more than 2 years? We are
still not ripe for maspar, no way, no Sir.
> > [ wired babies ]
>
> I think this would be ethical (if it could be done safely), since I
Of course. We are far from being able to do this now. Bioinert implanted
microelectrode arrays with no negative long-term effects are quite beyond
our reach.
> approve of all changes that increase the freedom of acting of a being
> (after all, the child doesn't have to use it if it doesn't want it, and
> is given new abilities). But I guess others disagree.
It would be interesting what the average extropian (a contradiction in
itself, obviously ;) thinks about this scheme?
> > [ inflating the neocortex by recombinant DNA tinkering ]
>
> Sounds quite possible, the brain gobbles up a surprisingly large fraction
> of our metabolism (was it 25%?). I think we would have to change the
Yes, I also seem to be recalling about this number. I am not sure, but I
think it involved a very high energetic load on the mother doing
embryonal development (caution, possible bullshit warning).
> development of the brain so that it grows further after birth (another
> bottleneck [and a sick pun]). Perhaps development should be slowed
Yeah, the birth channel. I never liked the amount of deformation
happening to a baby during birth. Uck.
> somewhat so that the brain can grow further, but I doubt parents could
> stand a longer neonatal period! (Unless we engineer their hormones
> too! :-)
Scotty, adjust the endorphine level by fifteen more units :)
> > [ lesion recovery ]
>
> This is what happens when the brain is injured. But I think the ass.
> cortices contain a lot of independent connections that doesn't correspond
> to any clear-cut regions.
'gene
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:35:43 MST