10 comments

[ 2.8 ms ] story [ 41.3 ms ] thread
Now only the Chinese pays for it...
Is it just me, but I find the name OpenCog terrible. Also it might be helpful to show 1 - 2 tutorials or examples to see what the "framework" actually does.
I tried compiling it last summer, and couldn't even get to the point of showing the dog model. The instructions for building and demoing really need some major updating.

Besides which, the scientific approach mainly seems to be just throwing together a whole bunch of sorta-kinda cognitive and perceptual algorithms with a unified knowledge base, and hoping something happens.

Last time I looked at it the Atom space db was single threaded and a nightmare to change to be non blocking. We tried.
Hi Jarrad -- FYI Linas Vepstas fixed the threading issues with the AtomSpace some time ago (over a year ago I think)....
As much as I love C++, I wonder if there has been any work on rewriting this in another language, even Java. From my experiences developing parallel versions of C++ and Java, most of my Java errors arose from program logic, while C++ arose from memory management. Also, thread management and building tend to be less cumbersome within other languages. For academic purposes, a language where you can focus on the concepts would seem to be more useful than a language chosen for speed.
Regarding the conceptual and theoretical background of the OpenCog approach, please see the two volumes "Engineering General Intelligence" published by Atlantis Press (distributed by Springer) a few months ago. There is a lot more subtlety to it than what you suggest.

Regarding the instructions for building and compiling OpenCog, it's a lot easier now than last summer. Try the Dockerfile, for example ... https://github.com/opencog/opencog/blob/master/Dockerfile

Ok, I will try to compile the dog again, and I might well queue Engineering General Intelligence onto my books-to-read list.
The project has really great ideas,the implementation tho is a mess. Like ubercode5 mentioned, C++ doesn't help their case. I think they really should bring it down to a very high language. Performance be damned. Just get it working first then optimize. Python/Lua will attract a lot of people. They can write the really CPU resource code in C and wrap em.
FYI, in OpenCog, MindAgents can be written in python or C++. Lots of OpenCog coding is done in Scheme these days as well.