Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Kuhn - relevence to computer chess -

Author: Joe Besogn

Date: 03:43:11 11/09/00

Go up one level in this thread


On November 09, 2000 at 06:34:34, Andrew Williams wrote:

>On November 09, 2000 at 05:20:39, Joe Besogn wrote:
>
>>On November 08, 2000 at 18:28:29, Andrew Williams wrote:
>>>To summarise: if you want to place yourself in the august company of Kuhn's
>>>scientific revolutionaries you first have to define what your "new paradigm"
>>>is. You could start by distinguishing clearly between the characteristics
>>>of an old-paradigm program and a new-paradigm program.
>>
>>From a post a couple of days ago:
>>If I had to pick one technical assumption of the old paradigm which is turned on
>>its head by the new - I'ld pick the concept of quiescence. The old paradigm says
>>search to quiescence, then evaluate the 'simple' quiet position, don't stop the
>>search in the unclear. The new paradigm says drive into the unclear and evaluate
>>with knowledge. Unclear is a good place for the new paradigm, it's a killer for
>>the old.
>>----------------------------------
>>
>>Clear enough for you? It seems very clear to me. But, pedantically ...
>>
>>a) at each node an evaluation function which contains the chess knowledge of the
>>programmer, I assume a reasonably competent chess playing programmer; in
>>particular knowledge of the king attack situation, with risky guesses as to
>>quantising king attack in terms of pawns; knowledge of all pins and all square
>>attacks, which pieces are attacked and with what danger, an attack-biased
>>evaluation which does not pay too much attention to defenders, the usual passed
>>pawn speculative and non-speculative code, etc.
>>
>>b) realisation that old-paradigm q-search is broken, fails to recognise unquiet
>>positions, is hopelessly geared to material only, has no concept of king
>>attacks.
>>
>>c) realisation that nullmove pruning within a materialistic search makes life
>>easier for a sacrifice attacking opponent, since the nullmove effectively
>>decreases the old-paradigm programs depth in the key sacrificial lines,
>>therefore reducing its ability to 'see' danger.
>>
>>d) realisation that the measurement system of the old paradigm is hopelessly
>>flawed.
>>
>>(i) that the programs play at tactical? 3000, and positional? 1900, and that
>>trying to express this as one number is silly.
>>
>>(ii) that rating system indications that programs are 2500 or 2600 and
>>competitive with GMs is silly.
>>
>>e) realisation that more and more search doesn't help against strong humans.
>>
>>f) realisation that with programs getting 'stronger', rating lists and rating
>>differences become irrelevant - style becomes important.
>>
>>b,c,d,e,f are old paradigm anomolies that are unable to be solved within that
>>framework of bigger-faster-better-materialism.
>>
>>a - knowledge and quantisation (rather than ignoring) of risky situations, fixes
>>to greater or lesser extents, the anomolies, including the major anomoly of the
>>q-search.
>>
>>The reason these are significant paradigm shifts is that the old paradigm
>>representative-in-chief says:
>>
>>1. q-search doesn't need to consider checks, nor does it need to be fixed to see
>>pins, because doing so renders the program slow, and the search explosive, and a
>>few errors never harmed anyone, since q-search is inherently error prone anyway.
>>
>>2. it is too risky being speculative, or applying scores to risky situations,
>>because you'll toss a piece and lose in the endgame. better to be 'accurate' or
>>as 'near accurate as possible'.
>>
>>new paradigm says: evaluate everything, especially risky situations, don't worry
>>about stopping the search when not quiet, since we trust our evaluation
>>function, in any case, quietness must consider king factors via the evaluation
>>function, accuracy is a false god, there's no such thing, guess.
>>
>>
>>Now, you seem to think that "open source" is required to discuss these issues.
>
>You have misunderstood. I was arguing *purely* about your identification with
>Kuhn. Kuhn was arguing with a tradition that said "Leave it to Science. Science
>is dispassionate and rational. The strongest ideas will always win out with
>Science because the scientific method, requiring full disclosure, will ensure
>it." Kuhn pointed out that the scientific method did *not* achieve everything
>that was claimed for it. My point was that if you're not reporting in detail
>everything you do, it's harder to complain if the tradition you are working
>in rejects your ideas. However, this post is an *excellent* description of
>your ideas. I'm sure I'm not alone in finding it interesting.
>
>>I think that is nuts. There is no black and white way to encapsulate the
>>knowledge, every programmer will do it his own way. There is no prescription for
>>evaluation values in risky situations, guess, each his own. Even with source,
>>all you eventually get are *conceptual* descriptions in your head using words.
>>Reading algorithmic source gives you no real *feel* of what happens. Listening
>>to the words of new paradigm programs designers works better, faster.
>>
>>I guess the new paradigm says "do it", quit trying to read blueprints.
>>
>>
>>Perhaps, en passant
>>>you could discuss whether an old-paradigm program can be converted to a new-
>>>paradigm program, or whether it really needs a new start from the ground up.
>>
>>Anyway you want. Everything is possible.
>>
>>>I'm not asking you to share any secrets here. if it's not possible to express
>>>it without giving too much away, so be it - but you could stop using Kuhn's
>>>views to insult people.
>>
>>Your's is the insult, this is an on-topic, non-insulting account which tries,
>>despite numerous attacks on it and attempts to get it to fly off sideways, to
>>stick to issues of Kuhn's ideas applicable to computer chess.
>>
>>Thank you.
>>
>>>
>
>If you want to believe that I am more insulting than you, you are welcome
>to believe that.
>
>Andrew



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.