Today I found out the OpenCog project

I usually step over the projects pretending to be the Next Big Thing(r) in AI, but this time I noticed that the project has been admitted to GSoC 2008 so, I thought, it couldn’t be so bad,could it?

Actually it looked very interesting and before I could even realize it, I was dwelling into the documentation…

It seems that the underling architecture is thoroughly documented but quite complicated. I’ll need some time to study it, but at a first glance they will provide a:

  • Knowledge Representation called AtomSpace (maybe similar to the one provided by OpenCyc)
  • Plugin container with the learning algorithms

The two things that captivated be were that they ship an improved version of the LinkGrammar parser that I used on my thesis and that they are working on a connection on OpenSimulator, a SecondLife open source reimplementation. Thus allowing the system to be embodied in a full 3D world…

I always thought that embodiment is fundamental to intelligence because having a body enables the intelligent agent to ground into reality all the words and the abstract categories it uses to reason. Notice that I didn’t used the word “program” but “agent” because I think that even humans use grounding to understand the words and the concepts they use day-by-day, but this will probably deserves a post on its own! 🙂

While waiting for intelligent robotic life to conquer the virtual worlds like Second Life, in the meantime I will watch one of my favourite films about AI: