This site exists to present a demonstration of the simulation of our automatic inference model. It includes a walk-through of an example case showing how our model does automatic inference. It also includes a runnable simulation that can run one of the demo inference networks we have created, or you can upload your own example network and see how inference proceeds.
Humans are able to think impressively quickly. In particular, even the simplest sort of language understanding requires a great deal of logical inference. When you hear the words "My dog Spot is a good dog," you probably infer that Spot has four legs and a wet nose and does not pee on the floor. However, you are not always conscious of how you arrived at those conclusions, and indeed, more reasoning is required than you probably expect.
Unfortunately, how this inference is done by your organic, neural brain is poorly understood. Not only must your brain infer properties, but it must keep these properties attached to the right objects. (For instance, hearing "My dog Spot is a good dog" should not make you think that the speaker has a cold, wet nose.) This is one aspect of the so-called "binding problem," which generally deals with the correct linking of properties. There have been a number of proposed models explaining how this can occur in the brain, ranging from holistic computation to localist models using a range of techniques.
One of the more popular models is a localist model using temporal synchrony called SHRUTI. It posits that there are clusters of neurons representing various concepts, and that the objects represented by these concepts are bound by neurons firing at the same time. Its explanations are impressively clear and plausible; however, it is not clear whether the brain actually uses temporal synchrony in this way.
In our technical report here (pdf), we propose an alternate model using "fluents" (short signatures, representing about 3 bits of information) instead of temporal synchrony. This model uses many of the same structures as SHRUTI but also introduces some new techniques, such as a method of unifying bindings. Like SHRUTI, clusters of tens to hundreds of neurons form nodes, which perform simple computations. These are linked together to build units which can represent belief in various propositions. A full description is included in the technical report; we encourage you to read that to understand our model.
The code for this simulation may be obtained here. Be warned that it has not been cleaned up for distribution, so it may still have references to predicates as focal clusters, among other things. For reference, of course, contact Leon Barrett.