On 6/26/19 6:03 PM, Basile Starynkevitch wrote:
On 6/26/19 5:21 PM, Stefan Reich via AGI wrote:
> Tying to build an AGI using symbolism, spoken language or C++ code
for example, IMO is like trying to build a car from cars. (I seem to
have a car theme going).
I beg to differ. I aim to create the "75% AGI" (everyday thought,
excluding just the highest levels of creativity) using symbolism.
And the bootstrapping approach, used since decades in the programming
language and compiler implementation domains, can and IMHO should be
transposable to such an AGI.
J.Pitrat is advocating such a bootstrapping and symbolic AGI approach
since several decades. See his blog on
http://bootstrappingartificialintelligence.fr/WordPress3/ and his
/Artificial Beings /book (unfortunately, that book is quite expensive,
but with some efforts you could find its PDF somewhere on the Web)
describing his CAIA
<http://jacques.pitrat.pagesperso-orange.fr/A%20Step%20toward%20an%20Artificial%20AI%20Scientist.pdf>
system.
I believe that, for practical purposes, /also/ using existing neuronal
AI and machine learning techniques and libraries would be worthwhile.
In other words, we would need (but that is very difficult and would
take more than a decade of full time research) to give symbolic
knowledge (actually meta-knowledge) to our AGI system about how to
cleverly use existing libraries such as TensorFlow
<https://www.tensorflow.org/> or Gudhi <http://gudhi.gforge.inria.fr/>
and that means how to cleverly generate some code using them (so
transpilation
<https://en.wikipedia.org/wiki/Source-to-source_compiler> techniques
are also relevant).
The basis (topmost layer, not by itself an AI by any means) of such as
system should probably be a /reflexive/, /multi-threaded/, nicely
interactive, and orthogonally persistent system capable of
meta-programming abilities, such as my Bismon
<https://github.com/bstarynk/bismon> (or some better near-equivalent
system).
Sorry for the typo. I meant *_bottom-most_ layer*, of course, similar to
Bismon for any such AGI system.
I am retiring in about 3 years, and I intend to devote a lot of my
retirement time to work on such an AGI (free software, on Linux)
system. I don't really expect to succeed, but I do expect to have a
lot of fun doing it.
If someone on this list is interesting in such an idea and accepts to
work on a free software (GPLv3+ licensed) Linux future AGI-related
system, please send me a private email (so reply to me only, not to
the list). But I don't think that a lot of readers on this AGI list
are as crazy as I am, so I would be really surprised if someone even
contacts me, since that AGI list is a "startup"-oriented list, not an
"academic" list to discuss implementation ideas and prototype software
systems. For personal ethical reasons (I am an excessively naive
idealist and consider software as something close to poetry, and as
beautiful as it), I don't even want to consider working on a
non-GPLv3+ AGI software. I have about software in general (and AI
systems) the same view as Erdös
<https://en.wikipedia.org/wiki/Paul_Erd%C5%91s> had about math and
proofs (when he spoke of "The Book" of theorems).
Also, I strongly believe that AGI is technically as difficult as
making a controlled-nuclear fusion
<https://en.wikipedia.org/wiki/Nuclear_fusion> reactor (consider the
ITER <https://en.wikipedia.org/wiki/ITER> and successors project for a
useful similarity, in terms of budgets, efforts, and time) or a human
space mission to Mars
<https://en.wikipedia.org/wiki/Human_mission_to_Mars>. There is no
simple and unique way to make AGI, the road is very difficult, and it
will take many decades (so I will be dead before seeing it) and
thousands of very clever people. I strongly believe that their cannot
be an "AGI in a garage" way of making an AGI. An AGI project should be
at least as ambitious and as big as the Manhattan project
<https://en.wikipedia.org/wiki/Manhattan_Project> was during WW2. For
software development organization of such software mega-projects
<https://en.wikipedia.org/wiki/Megaproject>, the SoftwareHeritage
<https://www.softwareheritage.org/> project may give a valuable
insight (see notably Roberto Di Cosmo <http://www.dicosmo.org/>'s
opinion on these). And that is why I believe that AGI has to be free
software.... Any proprietary approach will fail (because, like nuclear
fusion, it cannot be profitable in less than several decades).
Regards
--
Basile STARYNKEVITCH ==http://starynkevitch.net/Basile
opinions are mine only - les opinions sont seulement miennes
Bourg La Reine, France
*Artificial General Intelligence List
<https://agi.topicbox.com/latest>* / AGI / see discussions
<https://agi.topicbox.com/groups/agi> + participants
<https://agi.topicbox.com/groups/agi/members> + delivery options
<https://agi.topicbox.com/groups/agi/subscription> Permalink
<https://agi.topicbox.com/groups/agi/Tf97c751029c2e4db-M78fc68b73abbe92881a98845>
--
Basile STARYNKEVITCH == http://starynkevitch.net/Basile
opinions are mine only - les opinions sont seulement miennes
Bourg La Reine, France
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Tf97c751029c2e4db-M634b12436517e90edfd88cc6
Delivery options: https://agi.topicbox.com/groups/agi/subscription
