How can a mind watch itself? "Divide the brain tinto two parts, A and B. Connect the A-brain's inputs and outputs to the real world -- so it can sense what happens there. But don't connect the B-brain tot he outer world at all; instead, connect it so that the A-brain is the B-brain's world."
Now B can monitor and influence A. Why? to contorl, moderate, or adapt A to improve its performance. This arrangement is "reflective", and is partly "self-aware". (NB: It reminds me a little bit of Braitenberg's requirement in his later vehicles for monitoring overall activity in the vehicle's brain, so that thresholds could be continually adjusted. Perhaps for Braitenberg, the caudate nucleus is a kind of B-brain!)
"A B-brain could learn to play a role somewhat like taht of a counselor, psychologist, or management consultatn, who can assess a client's strategy without having to understand all the details of that client's profession. Without having any idea of what A's goals are, B might be albe to learn to tell when A is not accomplishing them but only going around in circles or wandering, confused because certain A-agents are repeating the same things over and over again."
In this design, some patterns of connectivity are unstable (e.g., reciprocal connections between A and B), eg. because of feedback driving the system out of control. Also in this design, many other layers of processors (C-Brains, D-Brains, etc. could be added -- e.g., to monitor the monitors.
(NB: From this, I get a sense of a multiple hierarchy, and it is difficult not to think of layers of hidden units in multilayer perceptrons. In PDP models, such additional layers are sources of computational power. Is this so for this idea of Minsky? Is the relationship between A- and B-brains nonlinear?)
Pearl Street | Society of Mind Home Page | Dawson Home Page |