RE: [computer-go] Pattern matching - example play (Quotes fromdeGroot)

["Persson, Magnus" <persson@xxxxxxxxxxxxxxxxxx>]

As a researcher in cognitive psychology I can affirm that Chrilly is correct in his interpretation of De Groot-thesis.
As a clarification I also think Vincent, who brought this up, was correct too but it was very easy to misunderstand what he wrote. I have learned that whenever Vincent writes something that sounds strange, one has to be very careful! He might be right...and often is (at least partially :-) ).

In some sense it is obvious that a good go players considers many more moves than a beginner does, but it is not this that matters.

What matters is that good players understand the position exactly and focus only on what matters. 

What we do not understand is what "knowledge and experience" (see quote below) means. Here are some very rough ideas of what might be going on inside go/chess masters brains.

a) A trend in current perception research is to view human vision as an optimal Baysian decision maker (that is the given the frequencies of given evidence, there is a way to combine the evidence in a close to optimal way). This would be almost similar to what Frank de Groot's statistical pattern matching approach does. 

b) It could also be something radically different. What if (now I am speculating wildly) practicing reading lead to independent small circuits in the brain that computes tactical results by local search. If this is done without awareness then a human is more like a huge cluster of highly gospecific processors. (This might also explain why GMs in chess cannot communicate chess knowledge as been shown earlier). 

c) There is a huge amount of high level (explicit or aware) knowledge that is perfectly tuned and bugfree (resembles what most goprogrammers try to do), and procedural knowledge how to apply this knowledge as well.

Perhaps all of this is true to some extent, but there are certainly some important things that are missing completely. I would also not be surprised if some (or all) of the ideas above are completely wrong. It could be the case that a and b is true and c in only an illusion provided by conscious thought. 

Whatever is true for the human mind does also not necessary constrain what is the best approach to computer go.

Theories in cognitive psychology (the stuff I work with daily as a researcher) are not helpful at all when it comes to truly understand the representation and processing of complex knowledge. We know what implications knowledge has for the ability to memorize random chess position, and we know that is likely that pattern recognition is an important process, but we do not know exactly how those patterns are learned, represented and processed.



-----Original Message-----
From: chrilly [mailto:chrilly@xxxxxxxxxxxxxxxxx] 

 
"Although a number of interesting findings on masters´ perception and memory werde documented in that study***, the results served primarily as evidence of Masters supererior playing strength - attributed  to their knowledge and experience rather than to their computational superiority".
 
Note: The kursive letters are directly from the original.
*** De Groots thesis from 1946: Het denken van de schaker. (Directly translated: The thought-process of the chess-player, the English title is the already mentioned "Thought and choice in chess").

--
Magnus Persson
Center for Adaptive Behavior and Cognition
Tel: +49-(30)-82406-350
Cell phone: +49 163 6639868
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/