Multiple Context: Blueprint for a Brain
The idea of a multiple context is that the many different areas of the brain bring together different collections of events, i.e. contexts, in the presence of which choices of action and predictions of stimulus can take place. This multiple context not only allows concurrent mental activity to progress in the fluid way that we experience in our thoughts, but at the more detailed level it allows for specific computational processes to occur. Much of my second book Associative Learning for a Robot Intelligence was devoted to examples of different arrangements of a multiple context for specific computational processes.
The multiple context system, PURR-PUSS (PP for short), is a design for a robot brain which learns like a human. Learning takes place by the association of actions and stimuli with contexts. The difference between learning in PP and classical associationism is that the associations in PP are discrete, i.e. digital, giving them the computational strength needed for language and other sequential tasks. The associations are interconnected in stochastic networks, which make the robot goal-seeking through a process called leakback. The PP design is for a parallel computing system, like the human brain, so my simulations of it on a serial computer are severely limited by the current speed of serial digital computers, impressive as they are.
One of my claims for a PP robot is that it will have free will. This is achieved by a simple device I call novelty. Every new association is marked as a goal until it is encountered again. These goals are selected by what is in the memory of PP, rather than being given by any designer, programmer or teacher, so they are PP's own goals. A surprise advantage of novelty-seeking over reward-seeking is that it makes PP more teachable. Further comments on the power of a multiple context together with some illustrative experimental results can be downloaded as the Word file MultipleContext.doc in the Downloads section.
The problem of consciousness continues to baffle everyone. Since we have failed to get animals to speak their minds, and there is a limit to the amount of surgical modification that can be carried out on a human brain, the robot may well prove to be the best means for testing theories of consciousness. To find out whether a brain of a particular design allows its owner to have a subjective experience, we will have to give a robot such a brain and ask it! Of course, this wouldn't enable us to know just what it was like for the robot, any more than we can know that for each other.
The idea behind the title of my first book--Thinking with the Teachable Machine-- was that the then primitive version of PP was useful for thinking about thinking. The current version of PP is even better for this. Designing a robot brain prevents one from being vague and allows one to see the results of new ideas. One day we may be able to study the effect of different brain designs on the subjective experience of their owners. Only the route through robots seems to offer us this possibility.