Roadmap for the Texai project

[Slippery Slope]

Hi Jim,

You asked "After several minutes on your site I can't figure out what
your approach is at a general level. Do you have a short intro
anywhere?".

In response I have posted a project roadmap at http://texai.org/blog/roadmap
. You will see that indeed my approach is at a general level, leading
to the Singularity. It is a disputed hypothesis however, that the
emphasis on symbolic logic and Good Old Fashioned AI will prove to
work. If I can get a compelling English dialog system going, then I
believe chances are improved that the remainder of the roadmap is
plausible.

Do you see any points of collaboration?

Cheers.
-Steve Reed

[Jim Lewis]

Hi Steve

Thanks for adding the road map to your site.

Your project certainly appears to be ambitious and I commend you for
such an undertaking.

My personal interests are more in narrower approaches, not in the
sense of narrow AI but in demonstrable advances in components of AGI
such as the more targeted work of Eric Baum and Robert Hecht-Nielsen.

I certainly encourage you to pursue whatever you feel comfortable with
although it seems to me that such broad approaches are overly
ambitious.

Perhaps over time, as this group develops, others will have more
useful comments on your plan and I am sure members will be interested
in any posts you make on your progress. Additionally you might get
more useful feedback by elaborating on a particular component of your
plan.

Onwards to the singularity!

Jim

[Stephen Reed]

Thanks for the encouragement!  Here is some background on the current Texai project phase.  Since leaving Cycorp in August 2006, I have been pursuing the goal of creating an intelligent dialogy system, along the lines of what Cycorp originally proposed for the DARPA RKF project http://reliant.teknowledge.com/RKF/proposals/Cycorp.html .  The Cyc Project, as perhaps you know, developed the world's largest and most comprehensive knowledge base.  They wanted to augment and elaborate the existing KB via dialog with subject matter experts.  What was delivered for the RKF project however, was much more narrowly tailored to the evaluation domains chosen by the DARPA progam and not really deployed by Cycorp beyond RKF.

I was the first RKF project manager for Cycorp and believe that I know how to create an intelligent dialog system according to Cycorp's original design with a few improvements of my own.  Some notable (existing, or yet to be written) facilities/components of the dialog system are:
* deep and precise understanding of the dialog user utterances, resulting in logical propositions
* discourse context maintenance (e.g. what are we talking about)
* user belief state maintenance (e.g. what has the user told the system)
* reversible grammar (e.g. anything that the system says, it should also understand if the user says it too)
* knowledge acquisition in the same manner as the Cycorp RKF systems
* skill acquisition beyond the current state-of-the-art

As the Texai Roadmap indicates, the dialog system is a critical component of the Albus (hierarchical control) Network that follows, enabling motivated humans to mentor Albus Nodes via English conversation.

Thanks in advance for comments.
-Steve

--
Stephen L. Reed

Artificial Intelligence Researcher
http://texai.org
3008 Oak Crest Ave.
Austin, Texas, USA 78704
512.791.7860

[Jim Lewis]

Hi Steve

I don't know that much about the Cyc project, frankly because it's
basic premise goes against everything I believe about how to do AGI.
AGI is about programming programming, not about programming in my
view. The concept of building a large knowledge base manually in fact
is entirely antithetical to what I believe is the correct approach -
programming knowledge acquisition instead of programming knowledge.
Perhaps I don't have the philosophy of the project right or am not up
to date and should read more about Cyc but unfortunately there's too
much to read and too little time -:(

Jim

[Stephen Reed]

Jim,

I too believe that it would be better to create a software application that can learn knowledge by being taught by a wider user base than programmers, logicians and philosophers.  That is my aim.  However I do know a great deal about Cyc, having worked at Cycorp for seven years.  I expect to manually create a bit more OpenCyc-compatible knowledge involving verbs ( i.e. situations, events, and relevant corresponding actors)  in order to  process the bootstrap sentence  vocabulary  of the initial Texai dialog system.  If my hypothesis is correct, given a sufficiently expressive bootstrap vocabulary, then the remaining lexical knowledge can be acquired via dialog.

For me reading about all the various (often narrow) approaches to AI (and AGI) is an important activity, which is a higher priority in general than coding.  It is very easy to progress too far down a development path that subsequently turns out to be less optimal than something that someone else has already written about.

Thanks for your comments.
-Steve

Jim

[michield]

Interesting project.
I see you are using Java. Have you considered mixing it with Scala?
http://en.wikipedia.org/wiki/Scala_programming_language

[Stephen Reed]

I am using Java because:

• I am very experienced in Java, especially non-GUI programming
• there exist a wide variety of libraries that I can include
• Java is mature, enterprise ready, and scalable to message-driven, Internet-sized computing
• the NetBeans IDE is ideal for rapid Java development and testing
• I am familiar with both object-oriented programming - having four years of Smalltalk background experience, and functional programming - having a seven years experience with Lisp.   Given my experience, I think Java is satisfactory.
  

Personally, I believe that AI can be created with neither choosing some new programming language, nor creating a new one.  I want very much for Texai to get to the point where it can program itself, and I think that eventually that will be CPU-dependent machine language - no source code, just bits.  Machine language programming and optimization will give a speed-up of maybe a factor of ten.

Cheers.
-Steve

--
Stephen L. Reed

Artificial Intelligence Researcher

http://texa.org/blog