Diagrams

[tkorrovi]

Diagrams from ai-forum posts

           I       I               I       I        
           I       I               I       I        
        ---------------        ---------------------
        I COMMON KNOT I        I OTHER COMMON KNOT I
        ---------------        ---------------------
           I       I               I    I           
           I       ----------------I--- I           
           I                       I  I I           
---------  I   ---------------------  I I  ---------
        I  I   I                      I I  I        
      ------------                   --------       
------I NEW KNOT I-------------------I KNOT I-------
      ------------                   --------       
        I    I                            I         
---------    I                            I         
             I                            I         
 -------------------             -------------------
 I NOT COMMON KNOT I             I NOT COMMON KNOT I
 -------------------             -------------------
    I           I                   I            I  
    I           I                   I            I  

[tkorrovi]

Triggers between two states

 -------------------------------------------------------------
 I       --------------------------------------------------- I
 I ------I--------------======= =======------------------  I I
 I I     I -------------I 131 I-I  35 I-------------    I  I I
 I I     I I       -----======= =======-----       I    I  I I
 I I     I I       I     I  I     I  I     I       I    I  I I
 I I     I I       I     I  ----  I  I     I       I    I  I I
------- ------- =======  I     I  I  I  ======= ======= =======
I 172 I I 158 I I 148 I  I  ---I---  I  I 143 I I 153 I I 165 I
------- ------- =======  I  I  I     I  ======= ======= =======
 I I     I I       I     I  I  ----  I     I    I I  I     I
 I I     I I       I     I  I     I  I     I    I I  I     I
 I I     I I       -----======= =======-----    I I  I     I
 I I     I -------------I  34 I-I 129 I---------- I  I     I
 I ------I--------------======= =======-----------I--I------
 I       ------------------------------------------  I
 -----------------------------------------------------

 -------------------------------------------------------------
 I       --------------------------------------------------- I
 I ------I--------------======= =======------------------  I I
 I I     I -------------I 131 I-I  35 I-------------    I  I I
 I I     I I       -----======= =======-----       I    I  I I
 I I     I I       I     I  I     I  I     I       I    I  I I
 I I     I I       I     I  ----  I  I     I       I    I  I I
------- ======= =======  I     I  I  I  ======= ------- -------
I 172 I I 148 I I 158 I  I  ---I---  I  I 143 I I 181 I I 194 I
------- ======= =======  I  I  I     I  ======= ------- -------
 I I     I I       I     I  I  ----  I     I    I I  I     I
 I I     I I       I     I  I     I  I     I    I I  I     I
 I I     I I       -----======= =======-----    I I  I     I
 I I     I -------------I  34 I-I 129 I---------- I  I     I
 I ------I--------------======= =======-----------I--I------
 I       ------------------------------------------  I
 -----------------------------------------------------

[tkorrovi]

Likely a similar structure would be necessary for repeating a sequence of letters. But you see that the start structure (provided in readme file) is very different from this, so it is unlikely that the test result (with di.str) is any inherent property of the start structure, it is rather completely a learned behaviour. Though the test is primitive, it needs a kind of prediction that output "hi " written in a single cycle would be followed by input "hi ". If it is so, then the system can predict its behaviour, and always writes "hi " (in a single cycle), assuming that input "hi " would follow. Now stop the system (uncheck "work" checkbox), change the training program to dt0.dll (an empty teaching function) and start the system again. As you see, the output is not always "hi " any more, but varies, it cannot act any more so that it can predict its behaviour, and tries to find alternatives, the most likely explanation so far. This is the only behaviour I trained it to, and I never tried to train it to do anything else.

The teaching program provides it its environment, in the simplest case it is just a reaction to its output, but it also could be like a simple command shell (like every command shell, it would act unexpectedly if something was done wrong). An important principle for the system which at first has almost no knowledge at all is that at first the environment must be very simple, and then it gradually must become more complicated. For example, I started with a reaction to much wider set of outputs (like "h " or "hi") and when in time it learned, and the correct output appeared, became more strict. Unfortunately such first stage of training "from zero" takes quite a long time, like few days working continuously for a modern PC. And unfortunately I never tried how much faster it becomes for a more learned system. We may certainly say that training such systems, any systems what start "from zero" (I hope you understand what almost absolute zero I mean here, almost no connections and certainly no reinforcement, not to speak about criterias in genetic algorithms or statistical analysis in dynamic neural networks), is quite slow, and quite complicated process. This may seem primitive, but understanding that process a bit more may have considerable theoretical importance.


( A proposed mechanism for AC implemented by a computer program: absolutely dynamic systems www.fortunecity.com/skyscraper/database/2 )

[zouze]

I am so amzed with what you are discussing here. This is a very nice forum.

[tkorrovi]

Thx. This forum is not exactly about Absolutely Dynamic System, it is about the theory to create an unrestricted AI system.