I want to develop a multinode distributed extensible application which can learn and communicate, with the ultimate goal of passing the Turing test. OpenAI would begin as an Open Source collaboration called Brian, composed of a client Neuron application and a distributed associative memory database on the client nodes. There will be numerous Neuron versions controlled through a permission and name based source control system, but all will use the underlying Neuron core and be able to communicate and share the distributed database.
My goal is to develop a large distributed associative database system which can be roamed and indexed by autonomous agents which represent independant thought processes. Learning and thinking agendas will be controlled in the client Neurons by selecting modular "personality" sets which can implement the algorithms desired by the client or by distributed "collective" processes.
As an Open Source initiative, all Neuron client code will be available. The initial implementation will be in 100% Java, but the protocols will be available to other implementations and we expect performance based platform specific versions versions to appear as needed.
The initial effort will be to install an emotion based pleasure and pain based feedback system into a distributed associative memory framework. Our hope is to build a motivational command and goal system on top of this emotional framework. At the same time, some of our contributors will be attempting to forcefeed the combined knowledge of all mankind into Brian's memory network. At some point we will then begin constructing thought agents which can attempt to operate on that memory data.
At this point, all of this is just a dream, but if you are interested in making this more than a dream, please join the OpenAI mailing list and help make this something real.