Re: Recycling/AI and super human computers

[T.L.G.vanderLinden@student.utwente.nl (Timothy van der Linden)]

Reply to Kelly:

>???!!  TV's have changed dramatically in the last thirty years.  Not the
>picture format, which is fixed by law, but the image clarity and relyability
>is far better.  That of course ignores the fact that current TV's are about
>to be phased out for digital HDTV's (at least in the states).

Yes, but untill HDTV (which is more or less the logical step to digital
multi media and communication) these improvements are not so big that every
living soul (overstatement) wants to buy a new television. In fact the step
to HDTV has to push a lot of unwilling people buying new TV's (unless their
old ones breaks down beyond repair.) Is the advantage of the Hifi audio and
video (probably compressed with MPEG (which hopefully is better than what I
see on my P5)) 
indeed that big that all people WANT to buy it?
My guess is that only new purposes like interactive TV, pay-TV, internet by
TV will tempt many people to buy such an apparatus.

>Medical gains are also accelerating, not declining.  Infectious deseases
>unfortunatly are also ready for a comback.

Yes, I read that that was because most the bacteria become resitant and
because the pharmaceutic <sp?> industry forgot that area but are now heading
for a come back.
 
>Virtually all major feilds from industrial architecture, computers,
>industrial manufacture, chemistry, aviation, farming, have all seen dramatic
>improvements.  Theses improvements don't always do something obvious, but
>they are there.

But are these advancements accelerations or just steady growth?

>> What happens if the jobs available are to difficult for less 
>> intelligent people.  After a while machines and AI will take 
>> over a lot of work. And if AI really become smarter than we 
>> are, then all the work we do would be superfluous.
>
>If the less intelegent can't find any job left at their skill level (or
>atleast not enough for all), they will eiather have to improve their
>abilities, or expect to be droped out of society.  Sooner or later people get
>feed up with others living off them for no good reason.  

Rethinking it, I come to the conclusion that as soon as AI has an IQ of 80
it will only take a few years to reach 180.

>As for us and super intelegent AI's.  Eiather we'll find something to do
>together (or co-evolve), or they will move on and ignore us. (We can hardly
>expect them to take care of us forever.)

Maybe we should make them just smart enough to do the dirty jobs. And use
only a few with IQ 1000 to think. Now we just have to hope they don't
interconnect without us noticing. Another possibility is to let them learn a
while and then freeze the learning capability. That would be like you were
stuck at a certain age, able to some specific job that needed little or no
adjustment.

>Best current bet is that the human brain has E10 Neurons and can process the
>equivelent of E14 bits per second.  Rough guess at computing power 1 teraflop
>(one trillion floating point calculations per secound).  

Yes, I found a single reference (after looking through 20 books about AI)
that confirmed your E10 (and not my E20) of information units (meaning
neurons*connections?).

>The biggest computer I know of personally is a 4-5 tera byte (E12 byte
>system) being constructed in my old neighborhood in Reston Virgina for the
>phone companies cable TV experement.  The Cray corporation is building a 9
>teraflop system for the US government DARPA research agency (completion
>scheduled for 1998), and their competitor (thinking machines) is offering a 2
>teraflop system comercially for $100 million. 

Yes, I found some info about another CRAY having max 1.2 TFLOPS.

>So we already are building systems bigger and more powerfull than the human
>brain, we just don't really know how to make them think.  True intelegence
>may require custom circutry.  But that circutry could be built in a way
>similar to the current circuts.  If of course it does need couston circuts,
>which is hotly debated.

Custom circuis are is less efficient to do highly parallel computing than
neural-circuits. So once the parallel computing is better controlled or once
the neural-activity is better understood the whole circuits will become much
more efficient. Of course non-chaotic system may always be more efficient on
less parallel circuits.

Timothy