On Sat, Mar 22, 2014 at 10:06 PM, Ben Goertzel <[email protected]> wrote:
> The goal with OpenCog is not to outdo CNNs or statistical MT on the
> particular problems for which they were developed.  The goal is to
> address general intelligence...

I realize that AGI should be able to do everything that the human
brain can do. I should be able to ask OpenCog to translate text to
another language, recognize faces in pictures, drive a car, or play
Jeopardy, and it would do it. I just wonder how you are addressing the
enormous computational resources (hardware, software, and training
data) needed to solve these problems. Certainly it must be a lot
harder to solve all of these problems at once than focus on just one.

The obvious application of AGI is automating human labor. The ROI over
15 years of world GDP should be about $1 quadrillion. I find it
curious that companies investing heavily in AI like Google, Facebook,
and IBM won't even invest $1 million in OpenCog. Are they really
setting odds of success at a billion to one against?

I realize there is a synergy effect in AGI. We do not just recognize
words or faces in isolation. We use context from all of our senses. So
it would seem justified to claim that we won't see any signs of
progress until all of the parts are finished. Then everything will
just work.

But regardless of whether I buy the argument, those with money don't
seem to. Surely there must be some way to indicate progress. The last
published experimental results or demo I am aware of is the 2009
virtual puppy video. https://www.youtube.com/watch?v=ii-qdubNsx0

The puppy video shows a synergy effect of combining a simple 3-D
virtual world, a simple (presumably rule based) language model, and a
simple 13 component model of physiological and emotional state. So it
would presumably be possible to periodically run system wide tests and
measure progress of a synergistic system. Are you doing this? What
results can you show to investors?

How do you plan to scale up the problem from one CPU to millions of
CPUs? I recall some tests on distributed AtomSpace that showed severe
performance problems. How do you plan to organize thousands of
software developers? How do you plan to collect petabytes of human
knowledge? How do you plan to acquire the computing power? Or do you
claim that solving all of the problems at once is easier than solving
one narrow-AI problem like Google translate or Watson? The last time I
checked, OpenCog is an (unfinished) Linux download I can run on a PC.

-- 
-- Matt Mahoney, [email protected]


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to