RE: Accelerating the seed AI development

From: Ben Goertzel (
Date: Mon Jul 30 2001 - 18:41:31 MDT

The main problem with opening Webmind development to a broad mix of
developers is

a) convincing developers that this project is worth working on

b) the amount of time it takes a developer to understand enough about WM to
be useful. While it is possible to spin small, separable problems off from
WM, nevertheless, the optimal solution to these problems can only
occasionally be defined without reference to the WM context in which the
problems are posed

Right now, WM is way past the "vision" stage and has led to loads of
difficult specific problems (mainly at the interfaces between different
modules (association-formation and concept formation, existential
quantifiers in higher-order inference and schema learning, etc.; but also
more basic stuff, like what's the best way to measure the "usefulness of
thinking about X recently", a quantity used in importance evaluation and
scheduling, etc. etc. etc.).

Furthermore, we're willing to open up the high-level conceptual docs to
anyone who's willing to sign a simple non-D document, and the codebase to
anyone who'll sign a non-D and who appears actually to be willing and able
to help with the work. This is not open-source, but it's open enough to
allow academics, for instance, to participate if they want to.

-- Ben G

  -----Original Message-----
  From: []On Behalf Of
Jack Richardson
  Sent: Monday, July 30, 2001 8:05 PM
  To: sl4
  Subject: Accelerating the seed AI development

  The points made by Ben and Eli regarding the work required to achieve a
human-level seed AI and then to build towards a super-human version
demonstrate the value of opening the process at the current stage to a
broader mix of developers.

  One way to do this without necessarily going all the way to reveal every
detail about the code is to identify specific problem areas where particular
kinds of algorithms may be useful. The first class minds on the main AI team
can describe the nature of the problem they are working to solve and provide
suggestions of possible algorithms. Then, hundreds of graduate level AI
folks could generate potential solutions where the most useful contributions
could then be explored in depth by the main research team.

  This is likely to be necessary as practical experience shows that the
dream of a couple thousand lines of code chugging away on Beowulf machines
don't get you very far. Without knowing the details of Ben's work, I can see
that he is running up against the many practical, technical problems that
require hard, grinding work to solve. Furthermore, there may be hundreds or
even thousands of these.

  We will know that progress is happening when Ben and Eli are raising lots
of specific problems that they are struggling to solve. Then, we'll know
they are past the vision and into the nuts and bolts.

  Good luck!

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT