??? Architecturally speaking the existence of multiple channels speeds the processing, by allowing multiple functionals to be operated in tandem, with only a relative speed loss associated with coordination of the multiple channels, in what way is this a random degeneration? Or am I being frozen out of this discussion by Jargon.... ;)

Random degeneration is a term from Quantum mechanics. A degeneration is a set of states with equal energy. It follows from symmetry of a physical system. The Random degeneration is a case when essentially different Quantum States have the same energy. Because the evolution of a physical system depends on initial state, the physical system can go on different ways with equal probability. Thus, one gets the mulichanneling.

I want to mark an important role of noise. in this case zone of the RD essentially is increased.

My objection to the term "Random degeneration" is twofold

1. simultaneity is not necessarily random, sometimes it is an optimal solution where two functions do not share dependencies, and

2. degeneration, is too loaded a term, for a context where there is simply multiple opportunities for optimization. Other than that, please do go on.

I didn't invent the term. Multichanneling in my consideration is a chance (fluctuation) from a physisist point of view. I consider the multichanneling as an important part of AI. Variety of the Universe follows from the multichanneling. As I understand you consider multichannel processing in computers. So, we have the different subjects.

Practical way for building adaptive optimization. (Multivariative regression analysis)

1. Take a target function and any number of parameters or their functional combinations.

2. On the database of experimental (monitored) results, calculate the parameters and Karunen-Loeve series eigenfunctions. You can use a svd() program on C++ or MathLab.

Two of main features of Karunen-Loeve expansion are their mutually orthogonality and uncorrelation of the expansion coefficients. Thus, at any number of parameters,their expansion leaves stable, and addition a parameter can't increase an error (including addition the same parameter).

3. Check convergence of the procedure by increasing the database.

4. Check the database on stationarity for different equal sets of samples.

5.Check influence of each parameter (icreasing an error) by excluding the parameter from the consideration.

6. Exclude the weakly influenting parameters.

Only Human Int. operation is a p.1

Well I'd have to object to that last statement about only human int. Operation if I can interpret it properly, since I am sure drosphila also have parallel processes.

What difference between Human and Artificial intelegence? On my opinion it is in shortage of knowlege about HI, itself and its functioning. Thus, we have two approaches (as in Physics) phenomenological (black box approach) and structural. I try to consider the first. You are right, Multicanneling is a feature any biological system because of capability to survive. It includes errand decisions. But HI distinguishes from the organic animal's one its capability to think in abstract categories, the main of them an image of a human of itself.

Well, that shortage, comes mostly from A.I. researchers not being willing to spend the time to get familiar with Neuroscience, and the fact that at the micro-biological layer we are still learning about the interface between the DNA and the environment. The whole realm of epigenetics has become significantly more important since we have learned that a cells dna strands interact with their environment during development, and as part of nerve plasticity.

Another factor I think, is the unwillingness of the A.I. crowd to move away from the computer model of memory, and towards an implicit model instead.

The final factor seems to be the failure of the psychology mavens to fully understand the nature of the tissues in data-management within the brain.

Dear All,

Firstly, I believe it is common knowledge that humans process information at several levels of abstraction. In addition, for those learning PROLOG, usually there is a intro chapter on Meta-Processing. In other areas such as Black-Board Reasoning there are similiar functionality

In order for the Artificial Intelligence (AI) to emulate HI (Human Intelligence),

these representation and processing of information or knowledge

need to be mapped or programmed into the AI system.

Anyway, it is not clear what is the business reason for the AI system to be built. Is it for commercial or government or military applications, or just pure RnD anything goes ?

Without knowing where we are going, the journey will be in-deterministic and infinite.

I hope this is useful

regards

Ramzani

I am most interested is seeing someone put these theories into a functional program. Anyone interested? Come on, I will work with you.

Of course I am interested Rich,

In digital hardware, I managed to do some early experiments in implicit memory using content addressable memory, and a logic gate I called a satysficing gate.

How would you simulate a content addressable memory, as a computer program?

I imagine we could figure out the logic for the satysficing gate, for any particular coding interval.

But to work best it should work for every particular coding interval.

Graeme, Have you put it in a robot? Have you seen some results? Do you have something you would like me to test in a robot? The final goal is to build a robot that can think for it's self. It should walk and talk like a person. Theory is continuing, but we need to get physical. Until it functions, and gets some kind of results, it is just more science fiction.

No Rich, my robotics at this moment are limited to ardweeny and stepper motors to create a 3D printer. However I have had some interest on that project from engineers so I might be able to develop some kind of a Robotics interest group, which will make the project more likely some time further down the road.

The problem is, that I need a specific type of program, and I am not creative enough to come up with it myself. Essentially I need to find a way to program content addressability, so that it eliminates the assumption of discrete addressing.

It can probably be designed as a sort of Virtual Machine, But my mind is still blocking on how to specify it. If I get away from the software logic, and look at it as a Bit Field, I might be able to come up with something, but, I need a general statement about how to handle the bit-field, and whenever I try to come up with that, logic gets in the way.