[FRIAM] singularity

[Bill Eldridge]

Carlos Gershenson wrote:
>> Tangentally, this question is part of the reason I am very disturbed
>> by the concept of the "singularity"
>>     
>
> I made yesterday a blog entry about the singularity:
> http://complexes.blogspot.com/2006/07/limits-of-moores-law.html
>
>   
Well, you note, " How the hell do you program a human mind in there???
It takes us several years just to learn to talk!"

Part of the answer is, "We just copy the ability from computer to computer".
Humans are difficult to clone. Machines much less so. Is it because machines
are so much less complex, or that the method nature chose was the best 
available
at the time or that human replication serves other purposes as well not 
satisfied
by an in-depth copy? A bit of all 3. Humans have not evolved terribly 
quickly,
but this model has had relatively long shelf life - much longer than an 
ENIAC, say.

As we digitize data, information and knowledge, it becomes easier to load
up a machine with it all. Obviously accessing and integrating this is 
more important
than just having it stored on relatively fast disk, but it's hard to 
deny that the ability
to store tons of knowledge is an advantage.

Machines have much faster data transfer internal-to-external.

Where humans do seem to win is in internal communications and the software
programming. I'm sure as we go to molecular computers we'll pick up some 
speed
on the internal bus bandwidth as well. Not that cognition is all about 
speed -
slow filtering is very useful in places.

Regarding the software, well, human development is a little bit stupid. 
Yes, it takes
us years to learn how to talk, and then we spend years learning "Row row 
row your boat"
and other time intensive learn-by-repetition-and-rote tasks just so we 
can be relatively
self-sufficient for 50 years, which means we hold meaningless jobs so we 
can find time
to head to the bar. But our external knowledge and technology cumulate,
so in 2016 we'll be able to organize computer knowledge much better than 
we do know,
and presumably as the machines get smarter and smarter, they can play a 
larger role
in programming their descendants.

The relevancy of Google answers is much better than we had 10 years ago, 
in a large
part due to comingling requests and answers over millions of nodes and 
requests, as well as the
algorithms that go into the responses. Where will this approach be in 10 
years?
What new insights, what new applications? Computers will be more capable of
aggregating insights from a billion more nodes and applying the insights 
to new problems.
While humans are getting better at programming the ability to have these 
insights,
we are not getting much better at having the insights ourselves. Our 
creative thinking
more and more depends on the machine for its completion.

That doesn't mean all computer questions are tackled with ease. There 
were big linguistics/
machine learning setbacks in the 1980's, AI was overhyped, etc. But 
these efforts don't so
much disappear as they recur as technological and societal environments 
become more
prepared to utilize them. Whether this all leads to a singularity 
followed by the Cyberiad,
or simply continues as a long-term symbiotic relationship (man and dog, 
computer and man),
I don't know - I favor the latter. But not because we can't program, 
only that the relationship
will continue to evolve in ways we find useful, and we've already made 
great progress on
what we'd like machines to do even in the short span of computer 
science. I don't expect
a single algorithm or insight to change everything - I imagine there 
will be a number of evolving,
slightly incompatible approaches, from which a few will gain sway and 
slowly be replaced.

In any case, just because software evolution has historically gone 
slower than hardware, I don't think
it's inherent in programming that that has to be true forever. For one 
thing, we've used hardware as
stable datum - program new tasks, but leave the hardware design 
consistent and backwards compatible.
So while the goal of the hardware is higher performance/efficiency, 
software has to have better performance
and more features. And attempts to improve automatic programming have 
had poor results. But
the state of programming is much improved over 1991, and my guess is 
that it's only a matter of time
before all of our efforts in different approaches hit something that 
pays off more exponentially.

Okay, I didn't address the one question - if you copy my mind out to 
disk and back into another body,
will it have identity as "Bill", self-knowledge, consciousness, etc.? I 
think Lem answered that in the Cyberiad,
but I'll have to re-read it, I don't store data that well.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/friam_redfish.com/attachments/20060719/e1fdc788/attachment-0001.html