The downside to chess

Originally posted by greenpawn34
It may appear like a wild sac game but everything would have been
calculated down to the smallest detail.[/b]

This is not true. There are plenty of computer games where a sacrifice has been played against another engine and the sacrifice has turned out to be insufficient.

When people remember how complex chess actually is, and realise that computers are often only scratching the surface, it stands to reason that even computers have to a some point stop calculating and evaluate the position. Modern engines will sometimes give more significance to king safety and piece activity over material count, and sometimes their evaluation is correct and sometimes it is not. Compared to humans, the computer's superior calculation will indeed help get it right more often, but it's still not 100% sound.

Originally posted by Marinkatomb

Fritz crunches all the available moves to 10 ply, no exceptions... Houdini doesn't do this. It disregards most of it's potential moves and focuses on 3-5 core first moves

Both Fritz and Houdini use alpha-beta pruning which is most common for top engines. With optimum move ordering, alpha-beta can reduce the branching factor in chess from an average of 35 moves to something like 6. And in doing so it has no risk of pruning an important move.

Robert Houdart, the author of Houdini, has openly acknowledged that he widely used ideas from other engines such as Ippolit and Stockfish (see the Houdini website). However there is controversial debate as to whether he actually used some of the source code - I have no opinion on this. So why is Houdini 3 so strong? Yes, he did improve the search but I don't believe he radically came up with a new approach. Instead I think the emphasis was mainly tuning: "Over 10 million chess games were played for the development and tuning of Houdini 3" (from his site).

it works it out again, going two ply deeper this time. This, when you are talking about computers, is inefficient.

This use of iterative deepening and the hashtable is fundamental to the alpha-beta algorithm and trying to achieve optimum move ordering - it's very efficient.

For others, as you probably know this, alpha-beta works fastest when it considers the best move first. But how do we know what is the best move since that's what we're trying to work out?! So it uses a shallower search initially since that uses less time, before doing a deeper search. Often the best moves in the shallow search remain the best when looking deeper.



Imagine calculating Kf1 20 moves deep, and then for f3, f4, g3, etc. before getting to Ra8 mate! You would have wasted a lot of time on those deep calculations. So consider all moves one move deep, realise immediately that Ra8 is best and note it in the hashtable. Then if you needed to look deeper (not in this example) you start with what is best in the hashtable. This allows you to skip redundant moves such as Kf1 (this is alpha-beta pruning in action).

Hi Varenka.

"There are plenty of computer games where a sacrifice has been played
against another engine and the sacrifice has turned out to be insufficient."

Is that not the horizon effect *- it looked good in 10 moves time but
on move 11 there is a huge hole. Next move the thing spots it's played
a lemon and does it's best to recover looking only at best moves.
It cannot play for complications, it does not know what complications are.

It will not attempt a swindle if there is a sounder move in the position.
Humans do this all the time.

* horizon effect:
I know the claim is this no longer affects computers but the thing must stop
analysing a position at some time , be happy with it and play a move.

Re computers making unsound sac v other computers.
Are you sure this was not a box using the old box psychology. 🙂

Originally posted by greenpawn34
Is that not the horizon effect

Earlier you were thinking "they are 100% sound" and "everything would have been calculated down to the smallest detail". Now you're talking in terms of "the horizon effect", "there is a huge hole" and "it's played a lemon". I agree more with the latter though it won't always be obvious for humans to spot.

The "horizon effect" did indeed plague earlier computers more. Due to having to play a move in e.g. 3 minutes, it would search lines to an equal depth of e.g. 10 moves. But humans (such as you GP!) know that you shouldn't stop calculating when there are further appealing captures, checks or threats to be considered. Early computers would stop calculating and try to evaluate positions with immediate tactics and get it wrong.... "oh, I'm a knight ahead! Actually no, he's capturing your queen on the next move".

Modern engines are much better in this respect. They'll keep calculating while there are checks, captures or threats (e.g. a passed pawn close to queening). They wait till the position becomes quiet and only then evaluate it. And I'm sure top programmers use even more advanced techniques that I'm not aware of.

Originally posted by Varenka
[b]Fritz crunches all the available moves to 10 ply, no exceptions... Houdini doesn't do this. It disregards most of it's potential moves and focuses on 3-5 core first moves

Both Fritz and Houdini use alpha-beta pruning which is most common for top engines. With optimum move ordering, alpha-beta can reduce the branching factor in chess from an averag ...[text shortened]... s allows you to skip redundant moves such as Kf1 (this is alpha-beta pruning in action).[/b]

Wow, it appears my knowledge of engine analysis is hopelessly out of date! LOL

Great post!

What do you think about the potential for engines to refine our approach to playing the game? It strikes me that programmers write algorithms with an expected out come, but often the engines start playing unexpectedly and often start giving the appearance of a certain style. The game i posted seems t give validity to a style human style that is often considered unsound (ie putting piece activity above material).

GM's obviously use engines to check their ideas, but do you foresee computers changing the way the over all game is played? (Rather than just the opening, as things stand..)

Originally posted by Marinkatomb
What do you think about the potential for engines to refine our approach to playing the game?

I'm not sure but I'll mention my limited experience. I currently play a game; analyse it myself; then with a GM via ICC/Skype; and only then with an engine.

Of course, there are many times when the GM and engine differ. Firstly, he's commenting on my moves with very little time per move and without total concentration while talking to me. But mainly it is because he is human. I'd say that I follow his advice 90% of the time because I think it's more correct and/or understandable.



As Black, I played Bxc3 here, after which the engine gives +0.11, so pretty much level. But the GM thought this exchange was much more of an error because he knows that Black will face a hard defence (which I failed to do and lost). He explained that a R+B are often better than a R+N, plus the play on both sides of the board suits the bishop more (things I knew but had been drowned out by my other thoughts about the position). But Houdini wasn't troubled by it as it can play accurately enough. So the engine didn't help my play in this case.



White to play. We know that passed pawn races can be important in endgames and it's often not a case of who has the most pawns as long as you promote first. I considered Kc2 with idea of Kb3/Kxb4 and supporting my pawns. But time is crucial and Black can interfere with my plans after Kc2 by playing Ra7 or Rc7+.

Here, Houdini's suggestion was easier to follow than the GM's. Houdini: a4!, bxa3, b4! +- since the king can walk to support the passed pawn via Ka2/Kxa3/etc. Fast and to the point. For me, an instructive endgame example of creating passed pawns as fast as possible.

So, I'm finding my play being improved by both engines and stronger players. Though, and GP will rejoice to hear, the GM is more often the source of advice I can take onboard. I don't forsee this changing anytime soon. GMs will always have a thought process that we can relate to more. But, I do find engines useful if we know what to dismiss and what to accept.

Originally posted by Varenka
[b]Fritz crunches all the available moves to 10 ply, no exceptions... Houdini doesn't do this. It disregards most of it's potential moves and focuses on 3-5 core first moves

Both Fritz and Houdini use alpha-beta pruning which is most common for top engines. With optimum move ordering, alpha-beta can reduce the branching factor in chess from an averag ...[text shortened]... s allows you to skip redundant moves such as Kf1 (this is alpha-beta pruning in action).[/b]

I am glad you posted all this. Generally, my impression is that people get all hung up on the raw calculating power of computers, and fail to appreciate what programmers bring to the table in terms of how to use that calculating power.

HIARCS is not Fritz, Fritz is not Houdini, and Houdini is not Shredder- and the programmers are the difference.

A computer has no way to evaluate a position at all, except whatever values are assigned by the programmer. Some are simple, such as checkmates, but it's getting to that point that makes the programming so interesting.

I was just messing about with a completed game.
In this position:



The computer came up with a beautiful move.

1. ... Bd4!



Just look at what this does! The rook is trapped, and white is in zugzwang!

The computer gives this line as best and winning for black:

2.Rg6



2. ... Bxg6 (??) 3.hxg6



It thinks black is won. It's actually a draw. I was 99.99% sure, but just to be sure, I checked with the endgame telebase.
The king is never leaving the corner.

I understand that engines better than mine have the telebases built in, but this is still an odd oversight.

Funnily enough, at first I was praising the beauty of Bd4, winning the exchange. Then, I decided to look a little deeper!

Hi Varenka.

Admit my knoweldge of up to date computers is zilch.
I recall playing 7 in a simul dislplay as an advertsing stunt for an American
company. Texas Instruments.
My picture and comments appeared in a 1980's computer magazine.
Adding something like the games were testing and I could not
relax till all the games were finished.
(actually not quite true as the games were pretty easy. I won 7-0)

It's good when you can pick a GM's or I'M's brain, though some can explain
better than others. Some just shrug their shoulders and give the impression
they are very good but don't know how.
That is a standard communcation problem.

Other's are brilliant, Jacob Aagaard and Paul Motwani are two that come
straight to mind. If you are mxing with good players some of it rubs off.

But you can pick up loads of wee tit bits from all kinds of players.
I think a well read 1800 player knows just as much about the game as a GM.
Some of the advice I've seen from on here and heard first hand from 1400 & 1500
players is rock solid sound.
The big difference is knowing what to do and actually having the skill to do it.

Back to computers, do you not think they are now so strong they are
useless to the average player. If spotting tactics is all they are for
then Fritz 6 is perfectly capable of doing that.
Anything deeper and you will have to be a very good player to separate the
the stuff that is worthwhile from the chaff.

And I'm sorry but playing over a computer v computer game does nothing
for me. I am not in the least impressed. That's a different game they are playing.
They don't have nerves , butterflies or doubts to overcome.

BTW I see for some reason you got a few thumbs down on the posts in this thread. Just wanted to add it was not me. I'm a thumbs up guy when I can bothered.

Originally posted by greenpawn34

The big difference is knowing what to do and actually having the skill to do it.

Agreed, and I believe you've read Rowson's "Chess for Zebras", so you'll know his thoughts on knowledge versus skill. I seldom find a stronger player telling me some new piece of knowledge. Instead, it's usually e.g. "why were you in a hurry to occupy the open file rather than improving your worst piece which is the knight on e2?". In the book by Soltis "What it Takes to Become a Chess Master", he's got a good chapter on "What Matters Most", essentially seeing the wood for the trees.

Back to computers, do you not think they are now so strong they are useless to the average player.

I know what you mean. When stepping through my games, I sometimes think about giving the engine another 30 seconds or so on a given position. Then I think "if the engine requires that amount of time to find something, I've got no hope of understanding it". So there's definitely the "complex calculation" type of moves where better engines fail to shed any light.

But there is also the strategic type of move that the engine requires lots of calculation to find, but a human may make more sense of it with a lot less calculation because we are better at thinking strategically. The latest Stockfish engine is about 4th best in the world. It doesn't use endgame tablebases and evaluates the following position as +8. Fritz 5.32 evaluates as +5.

White to move:



Houdini knows this is drawn because it uses tablebases, but it highlights the limitations of engine calculation (Stockfish was looking at 16 million positions per second for a minute and couldn't tell it was a draw).

Likewise, in my earlier example of creating a passed pawn with a4, bxa3, b4, this is not easy for engines. I doubt previous Fritz versions would find it. But yet, I could understand the play based on my human knowledge of passed pawn races.

I see for some reason you got a few thumbs down

🙂 A thumbs down is worth every word used to back it up. People who have something worthy to say, do so.

The amount of time that one needs to spend researching chess games is quite
unbelievable. I am presently playing 44 games and even though i have started with
the English in very single white game I am feeling quite overwhelmed at the amount of
research that one needs to do. I have a personal database of 4.5 million games and
every time a game branches out there are a new set of games and positions that i
need to consider to ascertain if its good for me. One cannot rely upon statistics. I
simply do not understand how people can cope with this amount, after this batch is
finished I plan to take a rest from chess, its physically and mentally exhausting.

Can't remember who first brought "emotion" into the discussion, but, It's both our downfall and savior in the chess game. Chess without the emotion is just time consuming.

Originally posted by Varenka

🙂 A thumbs down is worth every word used to back it up. People who have something worthy to say, do so.[/b]

Well said 🙂

Originally posted by robbie carrobie
The amount of time that one needs to spend researching chess games is quite
unbelievable. I am presently playing 44 games and even though i have started with
the English in very single white game I am feeling quite overwhelmed at the amount of
research that one needs to do. I have a personal database of 4.5 million games and
every time a ga ...[text shortened]... is batch is
finished I plan to take a rest from chess, its physically and mentally exhausting.

Why clog yourself up like this? I spend some time looking over Master games, but when i play i play moves that i think are best according to my own plans. Sure i check a database from time to time, or try to see if my move has been played before (just in case it has led to a quick loss) but if you spend too much time researching and trying to play a perfect game you are going to damage your OTB play. When all these aids are taken away, you need to rely on your own judgement. It is your ability to judge a position that is the thing you should be analysing. Leave the deep analysis till after the game..

Originally posted by Marinkatomb
Why clog yourself up like this? I spend some time looking over Master games, but when i play i play moves that i think are best according to my own plans. Sure i check a database from time to time, or try to see if my move has been played before (just in case it has led to a quick loss) but if you spend too much time researching and trying to play a perf ...[text shortened]... osition that is the thing you should be analysing. Leave the deep analysis till after the game..

I identify with this. RHP helped my OTB game at first, but as I have become more adept at the whole CC process ( I have even gone as far as doing "pawn chain" searches in chessbase, where I took the game and removed all the pieces, and then searched just based on the pawns), I think it has actually impaired my OTB mental process.

I am considering going "cold turkey" and playing my games in purely OTB style just to get back my mojo, and it's only my obligation to clan games that has held me back from immediate implementation.