RE: guaranteeing friendliness [more about reaching AGI now that Ben has improved the thread]

From: Herb Martin (HerbM@LearnQuick.Com)
Date: Sat Dec 03 2005 - 22:36:02 MST

From: Ben Goertzel

<snip past discussion>

> Well, you could look at Minsky's chapter in this book

Thank you. I was particularly amused by the review's
mention of HAL's "lip reading capabilities" reference the
findings that conventional speech recognition systems
can be improved by 10-50% through a visual model.

> or (when it finally comes out in a few months) my forthcoming edited
> volume on AGI,
> <,,4-147-22-43950079-0,00.html>

Ok, I will look forward to it.

Do you have any particular references (preferably online)
for "text summarization" -- that is, the ability to extract
summaries or to paraphrase text and information?

(This is a field that interests me greatly. Along with
the idea of programs that write and optimize programs
seems to be where much of the leverage can be found; and
where my own VERY modest efforts are directed.)

> > But, be warned that one possibility for this is that
> > 'someone' (like the NSA) is actively discouraging such
> > work and at the same time is offering strong incentives
> > for those who can and do pursue it to be "one of the
> > family" and keep such research under security veil.
> I know enough about both the AI field and the intel community to know
> that this is very unlikely to be true; but I realize there is no way
> to convince you that it isn't true, if you choose to maintain this
> belief.

Your testimony is evidence, a very useful data point,
and worthy of consideration, perhaps even tentative belief.

[It was only a possibility in any case -- it did happen
in the cryptography/cryptanalysis fields for many years.]

Personally I maintain few beliefs to a 100% level, but
must act as if that is the case (when the level is in
the high 9s) to operate actively and coherently in the
real world.

> > Al Qaeda? Or some rogue nation?
> No chance -- those countries have very little technological
> advancement

It was merely an example to cause thought -- India
also comes to mind, also Taiwan, Singapore, or China
are candidates -- not a claim that any of these would
be the developer, nor was it a suggestion for bookies
who wish to handicap the field.

> > It is unlikely that a "maverick breakthrough" will have much
> > effect until hardware increases by several orders of magnitude.
> And what evidence do you have for THIS assertion? As someone actively
> involved with AGI design, I am not nearly so sure as you are of
> this....

I don't have any strong evidence, except a general (weak)
belief that we are going to need to approach, maybe exceed,
the capacity of the human brain to reach a comparable
(human level) AI.

It is of course possible that optimizations (not possible
in our evolutionary constructed brains) will reduce these
requirements radically. It is also possible substitution for
neuro-transmitters and other non-connectionist human
hardware may increase the requirements.

[And evidence for such computer advantages includes the
clear fact that computer memories are much more accurate
than human memories for simple facts; also they are much
more accurate for pure calculations and even in symbolic
math they can bee much better than all but a rare few

> > And, those with the hardware will have the first chance
> > to actually try any such breakthroughs....
> You're making an unjustified assumption about the hardware
> requirements of a successful AGI design.

Yes, I was presuming a certain level of hardware beyond that
commonly available or even beyond that available on state of
the art machines such as Blue Gene.

> Based on my current estimates (which are based on the capabilities of
> my own Novamente AI system when run on a handful of machines), the
> minimum hardware requirements for an AGI are not *less* than a few
> hundred commodity PC's ... but I am not sure they are more than this.
> And a few hundred PC's are not all that expensive to network together.

You are in a much better position to know, and even 1000 PCs are
not a particularly difficult barrier.

If that will do it, then the most advanced machines already
have far more than the minimum hardware requirements.

We could probably setup a "SETI @ Home" like network and garner
a few thousand actively participating (at any one time) machines
if network delays would not be a serious problem to such

Note: I mean far more than 1000 PCs, but with at least this
many active consistently.

> -- Ben

I would love to hear more about how you see current (and near
future hardware) doing this....

This would move the singularity into the realm of "real soon
now", rather than 10-30 years in the future.

Herb Martin

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:54 MDT