-
   - >I agree that existing tests for human child intelligence are
   basically fine for AGIs being built and taught via a
   human-development-related path...


Yes, I'm talking about developmental psychology, language acquisition etc.,
it's not a dumb a/b/c test taking...

>However, I think such tests are useful as qualitative guides for folks
working on AGI. Â  I think if you start interpreting them as rigorous
>progress metrics, somebody is going to be able to hack a system to pass
those early childhood tests without any capability of extending further >on
to adult intelligence....

I'm talking about SIGI (Self-improving ...), it's a sensori-motor AGI that
lives and interact with environment, teachers and other agents (any other,
as general as a corresponding human child would be capable to, at least),
it's not a test-taking hack.

Such a SIGI doesn't have another way to learn than from raw sensory data
(initially), after some time it can learn and acquire quickly data with
more abstract forms of sensing and directly connect to high levels, but it
must have all initial sensory interfaces required to do human-like
activities which a human would do and could do if enter in its virtual
world, or if the machine is fed with data from the real world and interact
with the real world.

In order to be qualified as "3 year-old" the same system must be capable to
pass the requirements for all lower ages (kindergarten textbooks programs
are divided in weeks, and each week has different subjects, for earlier
ages one can pretend being a mother/parent).

The SIGI must start from a position similar to a human (with some
relaxations possible), e.g. at 0 years old it shouldn't be capable to speak
sentences, write, count from 1 to 20 etc.

.... Todor "Tosh" Arnaudov ....

.... Twenkid Research:  http://research.twenkid.com

.... Self-Improving General Intelligence Conference:
http://artificial-mind.blogspot.com/2012/07/news-sigi-2012-1-first-sigi-agi.html

.... Todor Arnaudov's Researches Blog: http://artificial-mind.blogspot.com


   - *From:* Ben Goertzel <[email protected]>
   - *To:* [email protected]
   - *Subject:* Re: [agi] Re: A test for less narrow artificial intelligence
   - *Date:* Wed, 26 Dec 2012 11:44:30 -0500

   I agree that existing tests for human child intelligence are basically
   fine for AGIs being built and taught via a human-development-related path...

   However, I think such tests are useful as qualitative guides for folks
   working on AGI. Â  I think if you start interpreting them as rigorous
   progress metrics, somebody is going to be able to hack a system to pass
   those early childhood tests without any capability of extending further on
   to adult intelligence.... Â

   -- Ben G



-------------------------------------------
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