IA/AI, soft vs. hard take off.

From: Max Comess (mcomess@ucla.edu)
Date: Fri May 03 2002 - 03:04:56 MDT


Will,

        I think that a lot of what you say is true, when you think about it
in chronological terms. Unless someone develops a seed/emergent/Friendly AI
on their own, and it initiates a hard take off, I believe we will see what
you describe. In fact, I believe that this process is already beginning and
that we should do every thing in our power to help it along, thus my desire
for a Singularity consortium.

However, I also believe that this process of intelligence amplification WILL
lead to a hard take off, because it will result in the creation of a seed
AI, either by emergence or design, by the growing power that results from
combining many of the less intelligent "unmotivated" AI subsystems(e.g. cyc,
Newsblaster, etc) on massive distributed PFlop level nets, such as Altnet.
This process will either create an emergent AI, or it will allow the
developers who use this system to create true AI. The only question is HOW
QUICKLY will these elements combine, and I believe the answer will be
determined by how it is developed. Such systems can either be left to form
on their own, in a somewhat linear accretive fashion, or they can be
actively combined in a way that will create as many exponentially accretive
forward feedback processes as possible. If we want the answer to be "AS
SOON AS POSSIBLE.", we should favor the latter. This is why we need a
consortium.

Exponential scenario: Altnet as it could be.

IF users are able to benefit directly from the increasing power of this
network, many more will join, faster, and it will become an accretive
accelerating process. If we want to grow this network and use it as a
platform to run IA, which will drive progress toward true AI, then it must
do several things:

                                1. Provide incentives and benefits for users
in order to fuel growth of network. I cannot think of a more palpable and
immediate benefit to joining
such a network then being able to use it's vast IA and AI resources.

                                2. Provide developers with a constantly
improving updated knowledge base that is run and managed by the very same IA
and AI software they are
developing!!! This will create an "intelligent platform" upon which futher
advances can be developed. It's just serving the same
function as a say, a code library, but instead of being a passive database
of code, it's an active intelligent platform that is continually being
improved upon, and is constantly running, as a whole.... on a PFlop level
system.

1 and 2 are essentially inseperable goals, because if we want to provide the
power to the network, we must provide a reason for people to supply that
power, and that reason is a powerful IA/AI system.

In order to achieve 1 and 2, it almost goes without saying that this must be
a *dedicated* platform, devoting all of it's considerable
resources towards AI and IA. Essentially we have to take all these dumb AI
and IA programs that run relatively slow at a few GFlops on individual
computers, and combine them and run them at ~PFlops on one massive
distributed computer. Therefore the only way to make this thing work is to
make sure that these AI and IA apps actually have PFlops avialable to them,
thus a large portion, if not all, of this computer's resources must be
dedicated to running them.

If these goals are implemented, then we should have a relatively powerful
and growing AI/IA system. Altnet is supposed to go online any day now, many
of these IA and AI subprograms already exist.... This can be implemented
NOW. If this system is created I believe that the Singularity will not be
long in coming, probably on the order of months.

If, however; we succumb to our baser tribal instincts:

        Linear Scenario: Altnet as currently envisioned.

KaZaa generates plenty of new subscribers all the time, and probably many of
them will join Altnet. However Altnet will not generate much interest, nor
should it, if it is used only as outsourced computing resources so that
companies can save money. Altnet will not grow exponentially because *it*
will have no reason to. *Its* only purpose is to make money, and this is the
only incentive provided to the user.

Therefore no outside projects. No selling to pharmas and biotechs! This has
to be determined right at the outset that the goal of Altnet is the
Singularity, not a quick profit for brilliantdigital. None of the magic can
happen if the PFlops aren't there to run the AI/IA.

----- Original Message -----
From: "Will Pearson" <w.pearson@mail.com>
To: <sl4@sysopmind.com>
Sent: Friday, May 03, 2002 5:36 AM
Subject: Leaving soon

> Anyway I think I don't have enough memes in common with this list to be
able to argue my points.
>
> I do believe that an increase in intelligence in the world will have a
major impact on what it means to be human. However I do not believe in the
hard take off. So even if you get a transhuman AI, unless it takes over the
whole internet, I don't believe it will go to super intelligence. Just my
beliefs, to explain why I am leaving. Here is my belief one last time, that
AI's without wills of their own interacting with humans controllers and each
other in a vast network of people will cause a meta-system transition/super
organism. Cross between web waking up and Intelligence augmentation.
>
> Good luck, I shall occaisionally look in the archives, and I get Kurweils
news letter, so undoubtably I will hear about the small breakthroughs you
make. If this movement makes the large breakthrough that you all hope for,
then the whole world will find out :)
>
> Will
> --
>
> _______________________________________________
> Sign-up for your own FREE Personalized E-mail at Mail.com
> http://www.mail.com/?sr=signup
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT