POLL:Man vs Machine ?

[Sedat Canbaz]

Dear Jean,

One thing more,if you will start to play vs with above chess engines

Then,please play against the engines with maximum performance
I mean without taking a piece handicap

Really i wonder a lot about what will be your performance

For any questions,please feel free to ask me

Best,
Sedat

[Uri Blass]

Part of the answer has to do with the time controls and why humans play better and better with more time. There is still the tactical vs positional play issue and we might now ask if computers generally outplay humans positionally.

Related to that comment is the fascinating subject of what strong humans do when they play chess with (say) 10 minutes to think per move. I'm trying to formulate an 'algorithm' for what goes on which is more detailed than what Kotov proposed back in the 70's.

If engines have an algorithm.....why can't humans?

When I made very rapid improvement in tournament chess I made up algorithms for myself. They were crude but they worked. At my level the primary problem was blunders, so I was determined to avoid blunders as much as possible and had several steps - one of involved doing the equivalent of a 1 ply search (which I did not do in time trouble) and I would ask myself what "good" is there in each legal move? Even the worst moves have something good about them if you use your imagination. This actually helped me play very imaginatively and I found tactics that were not obvious.

But the step that payed off the most was the last step. After getting a candidate move that I was ready to play, I had to convince myself that it was not a blunder (at least not an obvious blunder.) And I allocated a small amount of time checking the most obvious cases such is whether it would now be hanging, is the piece I'm moving defending something that is now vulnerable, can I be checkmated on the spot, etc. No matter how sure I was of the move before these steps I discovered that I still would benefit from doing this anyway. Most of this was trivial stuff, but the trivial will get you well beyond the median tournament player level.

I know this is not what you had in mind, but I think it would be fun to develop a deterministic move playing algorithm that you could "execute" in your brain. A human executed chess program. A computer could "test" the algorithm so you could work out the details. The only problem is that I would bet that you could not consistently execute the algorithm without making an occasional error. After a game you could check with a computer to see if you made any errors!

Such a human executed program would be very weak - but maybe it would still be over 1200 level. An enhanced version of such a program would incorporate some human judgement and might be much stronger.

Don,
I would LOVE to speak to more about this because I've also made quite a jump recently in elo and some of my ideas sound vaguely similar to yours.

A month ago I posted about a 'Residual Image' problem I was having. I've read Nikolai Krogius' book on this and he suggests asking the question: "What has changed on the board?" (i.e. to do with legal moves) after every 1 ply of search in your head. I tried it and....well it helped a lot.

Here's another idea I actually developed from a comment Vas made a couple of years ago:

Premise 1: Strong players (>2200 elo) play well and rarely blunder.
Premise 2: Strong players blunder regularly against top engines

False conclusion 1: Humans don't concentrate well against machines, they don't take the game seriously
Better conclusion 2: Super accurate play from the machine induces blunders from humans

Action taken by me: Develop an extensive repertoire that leads to complex positions (well known to me) that require accurate play.
Result observed: It's amazing how even strong players will quickly go astray. In a bad position they are prone to blundering if they under pressure.

I think that the question is what do you consider to be a blunder.

I think that a tactical mistake that the computer needs a big depth to find is not a blunder and I think that >2200 players rarely blunder by missing some tactics that the computer can find by 1 or 2 ply search.

[Uri Blass]

Maybe I can shed some light on this. I am 2350 elo, and I lose to computers over 90% of the time with a piece handicap at 5 minute blitz. Draws are not counted.

Hello dear Jean,

It sounds interesting...

When you have free time,can you play vs the bellow engines at 5 min

I wonder about we are using right starting Elo calculation or not

Code: Select all

174 Horizon 4.3.165 Beta         2404   58   56   110   59%  2340   19%
175 Flux 2.2                     2393   60   61   100   46%  2420   23%
176 Amy 0.8.7b t2                2388   54   54   120   45%  2429   23%
177 Gaia 3.5 x64                 2386   45   46   160   42%  2441   29%
178 Averno 0.81                  2383   54   53   116   59%  2320   29%
179 Resp 0.19                    2383   53   53   116   55%  2348   29%
180 Aice 0.99.2                  2383   60   60   100   54%  2354   17%
181 Tytan 9.32 x64 t2            2381   57   57   110   50%  2377   25%
182 Chezzz 1.0.3                 2379   52   52   120   54%  2356   31%
183 The Crazy Bishop 0052        2378   58   58   100   45%  2414   27%
184 BBChess 1.10                 2377   60   59   100   56%  2337   28%
185 Butcher 1.58 x64             2375   57   57   106   49%  2384   25%
186 Esc 1.16                     2373   58   57   110   59%  2309   18%
187 Queen 3.09                   2371   55   57   115   39%  2457   24%
188 Diablo 0.51 JA               2369   56   57   110   41%  2442   24%
189 Alfil 7.6                    2361   52   54   122   37%  2451   29%
190 EXchess 5.01 Beta            2355   59   60   100   45%  2385   26%
191 NanoSzachy 2.9               2351   62   61   100   57%  2298   16%
192 Popochin 3.0                 2350   55   55   110   50%  2352   27%
193 Arion 1.7                    2347   56   57   110   41%  2411   21%
194 Tornado 1.0 Mainz            2344   54   57   120   32%  2476   25%
195 Zeus 1.29                    2341   57   56   110   55%  2311   22%
196 Chispa 4.0.3                 2340   61   62   100   44%  2388   16%
197 RomiChess NG4                2339   54   56   120   34%  2454   23%
198 GreKo 5.4                    2334   53   54   124   42%  2383   23%
199 Asterisk v0.6                2334   57   56   104   55%  2300   30%
200 Ant 2006-F                   2328   49   50   136   40%  2397   29%
201 GES 1.36                     2327   56   57   110   44%  2366   25%
202 Anechka 0.08                 2310   59   60   100   41%  2374   22%
203 KnightX 1.92                 2309   58   59   104   44%  2353   20%
204 Rotor 0.2                    2304   54   55   115   43%  2353   27%
205 Sage 2.2a                    2303   59   59   100   48%  2317   24%
206 Scidlet 3.6                  2303   58   58   110   53%  2282   15%
207 Natwarlal 0.14               2303   57   58   110   45%  2342   16%
208 LittleThought 1.01 x64       2302   59   59   100   44%  2355   25%

Thanks,
Sedat

I believe that these engines are clearly stronger than 2350 even at long time control and not only at 5 minutes per game.

[EroSennin]

Ok I will try at some point to play against those engines. I can tell though that on my samsung 1.4 ghz smartphone I have scored 1/30 against stockfish without any handicaps.

[Sedat Canbaz]

Ok I will try at some point to play against those engines. I can tell though that on my samsung 1.4 ghz smartphone I have scored 1/30 against stockfish without any handicaps.

Many thanks for your interest...

Can you inform me please your hardware (processor in mhz),which are you planning to play vs the above engines

And what kind of opening book will use for the engines ?

Actually,i believe the current way is good idea...

I mean its not necessary exactly human players of 2800 Elo about comparing the strength of the current chess engines vs humans

And i think in this way,we can find the average real Elo difference between Engines vs Humans

I have no patience to see your results


Best,
Sedat

[Sedat Canbaz]

Maybe I can shed some light on this. I am 2350 elo, and I lose to computers over 90% of the time with a piece handicap at 5 minute blitz. Draws are not counted.

Hello dear Jean,

It sounds interesting...

When you have free time,can you play vs the bellow engines at 5 min

I wonder about we are using right starting Elo calculation or not

Code: Select all

174 Horizon 4.3.165 Beta         2404   58   56   110   59%  2340   19%
175 Flux 2.2                     2393   60   61   100   46%  2420   23%
176 Amy 0.8.7b t2                2388   54   54   120   45%  2429   23%
177 Gaia 3.5 x64                 2386   45   46   160   42%  2441   29%
178 Averno 0.81                  2383   54   53   116   59%  2320   29%
179 Resp 0.19                    2383   53   53   116   55%  2348   29%
180 Aice 0.99.2                  2383   60   60   100   54%  2354   17%
181 Tytan 9.32 x64 t2            2381   57   57   110   50%  2377   25%
182 Chezzz 1.0.3                 2379   52   52   120   54%  2356   31%
183 The Crazy Bishop 0052        2378   58   58   100   45%  2414   27%
184 BBChess 1.10                 2377   60   59   100   56%  2337   28%
185 Butcher 1.58 x64             2375   57   57   106   49%  2384   25%
186 Esc 1.16                     2373   58   57   110   59%  2309   18%
187 Queen 3.09                   2371   55   57   115   39%  2457   24%
188 Diablo 0.51 JA               2369   56   57   110   41%  2442   24%
189 Alfil 7.6                    2361   52   54   122   37%  2451   29%
190 EXchess 5.01 Beta            2355   59   60   100   45%  2385   26%
191 NanoSzachy 2.9               2351   62   61   100   57%  2298   16%
192 Popochin 3.0                 2350   55   55   110   50%  2352   27%
193 Arion 1.7                    2347   56   57   110   41%  2411   21%
194 Tornado 1.0 Mainz            2344   54   57   120   32%  2476   25%
195 Zeus 1.29                    2341   57   56   110   55%  2311   22%
196 Chispa 4.0.3                 2340   61   62   100   44%  2388   16%
197 RomiChess NG4                2339   54   56   120   34%  2454   23%
198 GreKo 5.4                    2334   53   54   124   42%  2383   23%
199 Asterisk v0.6                2334   57   56   104   55%  2300   30%
200 Ant 2006-F                   2328   49   50   136   40%  2397   29%
201 GES 1.36                     2327   56   57   110   44%  2366   25%
202 Anechka 0.08                 2310   59   60   100   41%  2374   22%
203 KnightX 1.92                 2309   58   59   104   44%  2353   20%
204 Rotor 0.2                    2304   54   55   115   43%  2353   27%
205 Sage 2.2a                    2303   59   59   100   48%  2317   24%
206 Scidlet 3.6                  2303   58   58   110   53%  2282   15%
207 Natwarlal 0.14               2303   57   58   110   45%  2342   16%
208 LittleThought 1.01 x64       2302   59   59   100   44%  2355   25%

Thanks,
Sedat

I believe that these engines are clearly stronger than 2350 even at long time control and not only at 5 minutes per game.

Thanks for the useful info dear Uri

Actually i picked some engines,which are scored average 2350 Elo:
http://sedatchess.110mb.com/index.php?p=1_43

I hope all this mystery will be solved in 2012

Best,
Sedat

[EroSennin]

With no opening book on my 1.86 ghz computer flux 2.2 achieved 19/21 against me. All games were 5 minute blitz.

[Sedat Canbaz]

With no opening book on my 1.86 ghz computer flux 2.2 achieved 19/21 against me. All games were 5 minute blitz.

First of all many thanks !!

Really,i am very surprised,honestly i did not expect a such score ?!

I think there is something wrong...,sure we need more games,but anyway almost +400 elo above your human rating ?!

Or it seems Flux 2.2 is a very strong anti-human master




For example,here are the ratings of Flux 2.2 (based on different rating list conditions)
CCRL (40/40):2441 Elo
SCCT (QX9650 @3.83 GHz, 5 min):2436 Elo
SCCT (Athlon 64 X2 4600, 20m+10s):2393 Elo
WBEC (40/40):2341 Elo
UEL (5 Min):2274 Elo
Flux 2.2's overall rating:2377 Elo

Btw,can you post the games


Best Regards,
Sedat

[Dr.Wael Deeb]

With no opening book on my 1.86 ghz computer flux 2.2 achieved 19/21 against me. All games were 5 minute blitz.

First of all many thanks !!

Really,i am very surprised,honestly i did not expect a such score ?!

I think there is something wrong...,sure we need more games,but anyway almost +400 elo above your human rating ?!

Or it seems Flux 2.2 is a very strong anti-human master




For example,here are the ratings of Flux 2.2 (based on different rating list conditions)
CCRL (40/40):2441 Elo
SCCT (QX9650 @3.83 GHz, 5 min):2436 Elo
SCCT (Athlon 64 X2 4600, 20m+10s):2393 Elo
WBEC (40/40):2341 Elo
UEL (5 Min):2274 Elo
Flux 2.2's overall rating:2377 Elo

Btw,can you post the games


Best Regards,
Sedat

You gotta be kidding me

Flux 2.2 is 2350 Elo in my private rating list after playing 224 games at long time controls....

I am simply amazed by this result
Dr.D

[EroSennin]

I played flux again. Now the timecontrol was 15 minutes. Flux won 7.5/10.