What happens when you cross open source with artificial intelligence? This is the grand experiment unfolding at the OpenCog project. The software’s authors are building a MindOS, designed to be broadly usable by researchers and developers. The first experiments with OpenCog are taking place in virtual worlds, controlling virtual agents, and outside virtual worlds with the comprehension of natural language. OpenCog project leader Ben Goertzel describes the convergence of the two in his recent post What do virtual worlds have to offer AI?

grid_globe

The OpenCog team is working to make life easier for AI developers, by building a flexible and reusable platform to handle systems-level work, and by designing standards for AI modules to obey in their interactions.  Just beginning its second year of development, the team is focused on systems-level work, experimenting with new AI modules, and supporting the first virtual world application developers working with a pre-release version of OpenCog.

OpenCog is a general AI platform within which a variety of different AI systems can be developed, ranging from relatively simple practical Ai applications to — the central focus of the platform — more ambitious projects aimed at complex learning and “artificial general intelligence.”

One OpenCog project in the works is RexPets — virtual pets operating in the realXtend virtual world.  Each pet has spontaneous behaviors guided by its individual personality, and generated by a combination of uncertain logic rules encoding the basic elements of dog behavior.  Pets can also learn new behaviors, via imitating their human teachers, and receiving reinforcement regarding how good their imitation is.

A significant amount of OpenCog’s initial code was donated by AI software firm Novamente LLC, and constitutes modified, open-sourced version of software objects that originated in the proprietary Novamente Cognition Engine software system.  The firm’s motivation for open-sourcing the code was to accelerate the development of powerful AI via bringing more programmers and more minds into the process that was possible within the confines of a small software firm.  The initial RexPets functionality, which should be available later this year, will have similarities to the functionality of Novamente’s virtual dogs in Second Life and Multiverse, some snapshots of which can be seen in these videos:

http://www.youtube.com/watch?v=AFYD7VpZI0I
http://www.youtube.com/watch?gl=BR&hl=pt&v=dr9aRzXFoDg
http://www.youtube.com/watch?v=g-oxn7XVnbY (also embedded below)

However, detailed plans exist to achieve much more sophisticated behaviors than these via integration of more advanced AI methods into OpenCog.   For instance, integrating OpenCog’s RelEx natural language processing system will allow virtual animals and other virtual agents to understand more complex language than these dogs currently can, and also generate simple language expressing their needs and observations, and answering questions.

Currently the most ambitious OpenCog project is OpenCogPrime, which is an attempt to create advanced artificial general intelligence at the human level and ultimately beyond. Based on years of research by OpenCog co-founder Dr. Ben Goertzel, and seeded by code and ideas from the Novamente Cognition Engine, OpenCogPrime is described in the wikibook at 

http://opencog.org/wiki/OpenCogPrime:WikiBook

which will be published in traditional paper form in 2010. OpenCogPrime incorporates the AI underlying the current virtual-worlds and natural language applications of OpenCog, along with many other aspects.  It is founded on a novel theory of “cognitive synergy”, which presents hypotheses regarding how the multiple components of an AI system must combine in order to emergently give rise to a generally intelligent system. The OpenCog framework presents an ideal playground for experimenting with such hypotheses.  

For the virtual world developer, OpenCog powers interactive AI characters and ambient AI characters, directing spontaneous behavior generation, pathfinding, obstacle negotiation, and decision making. OpenCog interfaces with virtual worlds and game engines using special proxies, custom written for each, allowing the OpenCog-powered AI character brains to run on their own dedicated servers.

Creating an open source framework for AI development is uncharted territory. Like building an entire operating system, it requires an overall vision and lots of hard work. Connecting AI systems to virtual worlds is also a new domain, as is the implementation of integrative AI designs like OpenCogPrime.  But it is precisely in this sort of area, where multiple adventures and uncertainties intersect, that there is the greatest possibility for radical progress.