Think Like a Computer

GP- It's from the first chapter of Tisdall's "Improve your chess now" - A very good book from the late 90's. I don't have my copy handy right now but I will try and look up the exact pages some time in the next couple days.

Cheers Nimzo.

I can get my hands on that book in a few days.
I'll have a read, it sounds pretty interesting.

Hi Scach

Yes I was talkiing about the strong engines. Fritz 6 and up.

I don't think I'm exaggerating when I say these things are brilliant at
tactics in 'heavy tactical position'

But perhaps my definition of 'heavy tactical position'needs clarifying.

I'm talking about real messy postions where tactics abound, not the 12-16
deep combinations where you are totally correct the 'horizon' can
catch them, especially with passed pawns.
(sometime they don't think it's threat till the pawn reaches the 6th or 7th rank.)

And again you are correct in saying they are counting points.
But it's not the 1,3,3,5,9 method. else they would never sac anything.
I'm sure it does not have a wee counter counting up the points off the board
like begineers do.

They are using a means to an end like chess players try to do and it would
be very interesting to know how they do it.
It must be something to do with how they evaluate a position which is often
open to question.

I'm not too sure there if there is anything we can learn from them in that
respect. Humans think and imagine differently.

How do you add up to 6?
I say 5+1, you say 3+3, another says 4+2 (Mad Rook says 3+4) 😉

A computer does not think or imagine.
It will have to look all 3 methods plus the 1,000's of other ways when it
adds 1.33 with 4.77.

I think it's method of evaluation (point counting) will be far to complex
for us to understand and grasp let alone trying to emulate it in a game
with a clock ticking.

Originally posted by greenpawn34
Cheers Nimzo.

I can get my hands on that book in a few days.
I'll have a read, it sounds pretty interesting.

Hi Scach

Yes I was talkiing about the strong engines. Fritz 6 and up.

I don't think I'm exaggerating when I say these things are brilliant at
tactics in 'heavy tactical position'

But perhaps my definition of 'heavy tactical positi ...[text shortened]... stand and grasp let alone trying to emulate it in a game
with a clock ticking.

How do I add up to 6? Man, that higher math gets me every time!

@Varenka, although your thread is entertaining, it also opened the door for Greenpawn to go into his "boxes got no intuition" rant. 🙂

@Greenpawn: I'm not quite sure how to respond to the thread in general. Two large parts of chess engines are the search function and the evaluation function. Many parts of the engine's search function are done in a completely non-human way. (The Kotov tree versus the De Groot/Tisdall discussion.)

However, most parts of the evaluation function are modeled in a completely human way. (Or at least, that's the goal.) The piece values are a large part of the evaluation function, but the evaluation function is also typically composed of many more positional factors. (Development, board control, mobility, king safety, pawn structures, etc.) And these factors sometimes change depending on the phase of the game. By the way, concerning piece values, most of the engines seem to be using the Kaufman values these days (1, 3.25, 3.25, 5, 9.75, with a 1/2 pawn advantage for the bishop pair) instead of the older Reinfeld values (1, 3, 3, 5, 9), although I've seen some values even different from those two sets.

Regarding the intuition argument, I guess you're right about that. Engines seem to have a hard time acting on a hunch. Although some engines do have some kind of "aggressiveness" setting that supposedly allows them to take more chances. But I have no idea how that feature works.

GP, off the top of your head, can you think of a few good examples of human intuitive sacs? I tried a number of engines on the Fischer-Byrne "game of the century" (Fischer's 17...Be6) , but all of the engines I tried found Fischer's move almost instantly. Maybe that move is highly overrated for its creativity?

I think the original poster would find forcing chess moves a fantastic book.

Hi MR.

If you look at most of the big famous sacrifices and the players are
honest in their notes they will tell you it was based on pure intuition.

Google: 'Alekhine + big nose' not a joke Mad Rook. google it.

And Tal sacced on intuition at every chance
(though bad example as he was totally gifted).

Look at the Kaspaove game in Thread 133902 he never
worked out the whole game.

He saw counter play and an actve position.
The Queen sac was intuition giving a human player a very difficult
position to play. Computer s have such concept, how do they know
what a human finds difficult to play?

Put the box on the Kasparov game , does it play the sac.
(does it consider the sac? I don't know).

The Fischer 'Game of the century' is not really a test.

To be honest it's was blown up out of all proportion and only praised
because of Fischer's age.

Fischer himself left it out of his 60 Memorable games so it shows you
what he thought of it.

PS:
I don't think I've had a rant. I think I am rather mild this time.
We are discussing if we can learn anything from them at all.

I've always said yes. It can help, it's a tool and used correctly
in the hands of good player there is no doubt it can assist.

It's when the claims that playing the things actually improves
your chess is when I rant.

We produced some brilliant chess players from Morphy to Kasparov
without their help. Fact.

If playing the things makes you improve then how come the world
is not simply bursting with IM's and GM's?
People have been playing these things for over 20 years.
Where are all the new GM's and IM's there should be millions of them.

That's a rant.

Now google. 'Alekhine + big nose'

Originally posted by greenpawn34
Hi MR.

If you look at most of the big famous sacrifices and the players are
honest in their notes they will tell you it was based on pure intuition.

Google: 'Alekhine + big nose' not a joke Mad Rook. google it.

And Tal sacced on intuition at every chance
(though bad example as he was totally gifted).

Look at the Kaspaove game in [threadid ...[text shortened]... d be millions of them.

That's a rant.

Now google. 'Alekhine + big nose'

For what it's worth, all of the engines I tried liked Alekhine's rook sac by a slight margin, usually taking one or two seconds to find. None of the engines wanted to get anywhere close to Kasparov's queen sac.

If computers could teach us how to get a big nose then
they might on to something.

Pinocchio must have been a great chess player.

If someone studied the subject of flying, they'd probably study how birds fly; how aeroplanes fly; etc. Their understanding of what is fundamental to flight would be increased by comparing the various approaches. But still, they wouldn't expect birds to stop flapping their wings and instead start pretending to be an aeroplane.

Computer chess started around about the 1970s or so. By that point, humans had studied chess for a long time and we were aware of many principles. But which of these would need to transcend into computer chess in order to make them play well? Or would machines be able to use alternatives that would make the human principle unnecessary for them (just like aeroplanes not needing to flap their wings)?

In my original post, I tried to highlight things that humans do which computer chess has also shown to be necessary (so far). To me, this helps highlight what may be fundamental to chess. It's like comparing Capablanca and Tal and while realising their approach to chess was different, there are of course things that all top human players share and this overlap is interesting. But again, we're not saying Tal should have played like Capablanca or vice versa.

So please forget the ambiguous title and note my initial "we cannot or should not try to emulate - afterall, our brains work very differently. However, some aspects are interesting to compare and may even be instructive…".

Let's compare humans and computers in mathematics.

A human and a calculator can both tell you 2 times 2 = 4 in a second.

Now, lets try 244,246 times 34,467. A calculator will find this too in a second.

A human (or at least most) won't be able to keep up.

That's what happens in most chess games. There just isn't enough time or energy for a human to compete with this.

It doesn't matter whether the computer is using the same formula or has worked out some shortcut (unlikely I know).
Knowing how the computer does it will not allow humans to catch up, in my opinion.

Originally posted by paulbuchmanfromfics
Knowing how the computer does it will not allow humans to catch up

It's not about catching up anymore than it's getting birds to fly like aeroplanes! 🙂

Suppose that all world champions significantly used approach 'X' in their play, while only one world champion used approach 'Y'. From that, which approach sounds most fundamental to you? By highlighting the things that all top players do - even computers - we can add weight to well known principles. But I'm not saying we have to implement them like a computer.

Likewise, there are many aspects to Kasparov's thinking that neither you nor I could ever come close to repeating, but it's still instructive to consider how he arrives at his moves.

Nobody teaches birds how to fly, they just do it.

If this is related to how chess computers can help humans,
we have to ask ourselves what bird has learned to fly better
after looking at an areoplane.

Have you ever see a pigeon skooshing along the street with flames coming
out it's ass going over 600mph?

Originally posted by Mad Rook
For what it's worth, all of the engines I tried liked Alekhine's rook sac by a slight margin, usually taking one or two seconds to find. None of the engines wanted to get anywhere close to Kasparov's queen sac.

Happy news, I found an engine that will play Kasparov's queen sac.

Ed Schroder's Pro Deo 1.6. You have to pick the Michael Tal personality AND set the playing style to either active or aggressive.

On the active setting, it chooses the 12...Nfxe4 after about two minutes.

On the aggressive setting, it chooses 12...Nfxe4 after about 14 seconds, then at about the 36-second point, it switches over to 12...Nfxd5. Then at about the 3:20 point, it switches back to 12...Nfxe4.

Pro Deo is a free engine, but it's a Winboard engine, and it's a bit difficult to install (read the installation instructions).
Link to Schroder's site (Pro Deo 1.6 link is on the right side of the page):

http://www.top-5000.nl/

P.S. I've also heard that the commercial Junior engine can also play speculative sacs, but I've never used that engine.

Computers don't think, human programmers do. Humans take so many things in their thinking processes for granted -- they're wholistic -- whereas EVERYTHING has to be spelled out for computers, and even then they remain STUPID: no consciousness, no judgement, no perception. The ILLUSION of these things must be programmed in by humans.

In chess engine programming, the problem combines these factors: the "rules" which human experts use are fluid, not always explicit, are highly contextual and subject to modification, and are difficult to render "completely" into fixed algorithms. That's why it takes so much work, and so much artificial and odd methodology, in programming, to get a decent approximation of this.

But it isn't a case of "computers thinking differently than humans" -- since computers don't think at all. It's a case of human computer programmers attempting to distill the thinking of chess experts -- not always overlapping with the thinking of computer programming experts -- into a set of highly peculiar, redundant, and not very efficient instructions and loops of instructions for a computer to carry out while acting as a chess engine. The question isn't one of how computers think -- it's one of how human programmers think in trying to implement thought processes in terms of non-sentient, inflexible formulae as carried out by a non-thinking machine which is itself without understanding or judgment.

Greenpawn, however you define "real messy positions" they must of necessity be LESS complex and/or lengthy than sequences running up against the chess engine's tactical horizon.

I would also suggest that this horizon is less fixed than is sometimes imagined: as I wrote earlier, "Even chess engines can't examine all lines and don't try to, because they are working under time constraints; heuristics determine their move search parameters, and time constraints give them a tactical horizon beyond which number of moves in a given line they do not search. These do not always produce perfect results even in heavily tactical positions."

So, the "heavier" the tactical possibilities the more the search lines must be pruned and the more that time limitations must artificially (and sometimes catastrophically) eat into computer analysis. For this reason, I suggest that it is more sensible to talk about a window wherein a particular chess engine is able to deal with tactical lines better (on average) than a particular human opponent. Very simple lines may not give the computer any tactical advantage; but highly complex lines may not either, especially if the human opponent is advanced and the time controls are such as to provide adequate analysis time for the human player.

Edit: P.S. I think this is why so many computer opponents "prefer" rapid chess. 😉

P.S. We REALLY get the worst of both worlds when computer programs are given "personalities" because whereas chess, resistant though it is to algorithmic programming, is actually comparatively simple to get sense from relative to personalities. Chess is a game played from a simple set of limited rules which are themselves algorithmic, though their development into principles of play by human players is considerably less so; whereas personalities are something intrinsically non-algorithmic and even the simplest errors stand out as something unutterably alien. Chess judgment can be mimicked tolerably well by a computer, whereas personality judgment cannot because computers have no judgment and personalities are entirely judgment (not rule) dependent.

Originally posted by Schach Attack
P.S. We REALLY get the worst of both worlds when computer programs are given "personalities" because whereas chess, resistant though it is to algorithmic programming, is actually comparatively simple to get sense from relative to personalities. Chess is a game played from a simple set of limited rules which are themselves algorithmic, though their dev ...[text shortened]... cause computers have no judgment and personalities are entirely judgment (not rule) dependent.

Well, I guess I'm not much of a believer in various engine personalities like Kramnik, Anand, Fischer, etc. (Mainly because I don't think I could tell the difference, and I suspect it's more of a gimmick anyway.)

Anywho, I *am* having fun playing with the Pro Deo Tal personality. It's pretty weak, around 2000 elo. I edited my own version and got it to play at around 2400 to 2500, while still making speculative sacs. I was thumbing through Rampant Chess, looking for speculative sacs, and I was amazed that the Tal personality played most of them. If I had the time, I'd go through some of Tal's games to see how many of those the engine will play.