Re: [computer-go] Pattern matching - example play

["Frank de Groot" <frank@xxxxxxxxxxxxx>]

----- Original Message ----- 
From: "Don Dailey" <drd@xxxxxxxxxxxxxxxxx>
Subject: Re: [computer-go] Pattern matching - example play



> I  guess  what I'm  saying  is that  good  players  mainly gets  their
> knowledge from explanations, by a teacher, by being taught.

Has that been demonstrated?
It seems counter-intuitive to me.
I would think that Go players mainly learn from playing games against
stronger players.

Just as (most?) people learn a language best by immersion, practice and
exposure, and not from studying grammar books.

> lessons  by  the  individual  "students"  are learned  more  from  the
> ACCUMULATION  of knowledge  and  experience passed  down to  students.

Isn't half a million games "an "accumulation" of all previous Go knowledge
condensed into those games?


> Relativity in physics  is an example.  I doubt  very many people would
> have  "learned" relativity  from direct  physical observation,  it was
> enough that one man did.

Relativity is a theory which has not yet been proven satisfactorily.
Just the fact that it helps predict and "explains" certain previously
ill-understood phenomena does not mean the theory is correct.
In fact I reject the special theory of relativity. Einstein's own
explanation of it clearly is bogus, as well as the classic experiments that
are supposed to confirm it.

For ex. when they put an atomic clock in SkyLab and demonstrated that time
slowed down as predicted by the Relativity Theory, I immediately thought:
"But this is nonsense, they just demonstrated that atoms resonate slower
when you centrifuge them".
We forget that atomic clocks measure the resonance of atoms, they do not
measure time!
Time did not slow down, but when you subject an atom to a centrifugal force,
it resonates a bit slower.

My point is that all dogma's usually are false or at best incomplete.
The relativity theory "proved" that Newtonion physics was incomplete, and
one day someone will show that "special relativity" is incomplete.

This is why I would prefer a learning system that can re-learn when Go
"science" has progressed.
The learning system would forget the dogma of not starting at Tengen and
learn that starting at Tengen is better than not starting at Tengen, etc.
(to give a theoretical example).

I don't care much for dogma's, in fact I am pretty sure that just about
anything currently accepted in science is preposterous nonsense, and in Go
as well.

I think human Go pro's play Go very sub-optimally so learning from pro games
will never produce anything "better" than a Go pro, *but it will keep pace*
with pro patterns, for example.

The nice thing about learning systems is that they KEEP PACE WITH CURRENT
KNOWLEDGE.

When you are 20 years older and your Go patterns are a bit outdated, you
don't want to bother making a new set, you want to feed your system perhaps
2 million pro records (when they have finally become available..).


> I could  compare what you are  doing to trying  to learn sophisticated
> science without  the benefit of an  actual teacher but  it's not quite
> that  simple.  You are  trying to  teach from  GOOD examples,  which I
> consider a big step higher  than learning from scratch, but still more
> or less indirect.

Most absolutely not.

Listen.
When I learn form a pro game, I am not learning from good examples ONLY.
When a pro does a move, I have a "good" example.
But **all the other moves are worse or at best equally good (ON AVERAGE)**

Meaning, I have many more bad examples than good examples!

My pattern system very much uses this knowledge!

It knows that this and that and such patterns are rarely played on compared
to others because they are "bad", as a pro considered them "bad" :)

Having confirmed "good" examples automatically means you have the bad
examples, as bad = not(good).
Q.E.D.


> I'm not  trying to discourage  you on this,

Don't worry nobody can discourage me.
I feed on critiscism in fact.
The best way to make me do something is tell me "It can't be done".
And I am my own biggest criticaster.
When I get some better results I always tell myself: "You pathetic moron,
this is useless garbage, now throw that crap away and start all over again
but BETTER this time".
(Has something to do with my upbringing :)

(as my improved pattern system is still not even 1/3 with learning, it
already has better results than published on my site).


> you are doing  some great
> things here

I really am not soliciting for praise, in fact that rather makes me shy to
speak about what I am doing.
I really do not need any encouragement.
I much more prefer strong and hard critiscism (but with logical arguments,
not religious beliefs or petty personal attacks).
Only bone-hard critiscism with strong arguments bring a science forward.
Like Nils Bohr to Einstein: "Who are you to prescribe to God that he should
not play dice" (paraphrase).
People love their dogma's and I think almost all dogma's are just baloney.

I don't play Go but I am pretty sure that the game should be started at
Tengen.
The fact that only 0.3% does it (isn't a pattern system nice :) is evidence
to me that even pro's love their dogma's.


>  that will  provide more insight  into these  things.  For
> instance, I  admit that I could be  wrong about this (I  would like to
> know if I  am) and you could make the point  that everything we learn,
> even if  carefully explained to  us, is ultimately learned  by example
> (the example is "spoon fed" into our imagination when explained.)

When a Go player learns by a spoonfed dogma, that dogma comes from another
player that either reiterates that dogma, or has learnt it from playing Go
and observing Go and combining other dogma's that have a similar etiology.


>    Meaning, the programmer is an extra link in the chain.
>
> Again, are you saying that a "teacher" is a waste of time?   The
programmer
> is a teacher.


No, I just say that I would say that learning from games is "first hand
knowledge" and a teacher is "second hand knowledge".
The programmer as a teacher should IMO limit himself to teaching the rules
and a few other things that are 100% "certain" and easy to implement by
programming.


> Let's put  hero worship aside, even  the best players in  the world at
> any game  have their idiosyncracies  and even dare I  say "weaknesses"
> and in the end you would want to avoid that.

Yes, viz. above my remark that you occasionaly re-train certain aspects to
remain state-of-the-art.
This way you would still be able to play better than pro's because you would
have their combined knwoledge PLUS ever-so-many-MIPS.


> Is there any  way your system could generate it's  own games, and feed
> back on  itself, gradually learning  from it's own  experiences?

Most certainly not :)


> This
> would actually at  least introduce the possibility of  it going beyond
> it's teacher.


I am a firm believer in search :)

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/