Author: Robert Hyatt
Date: 19:15:36 08/23/99
Go up one level in this thread
On August 23, 1999 at 20:17:12, James Robertson wrote:
>On August 23, 1999 at 19:58:11, Andrew Dados wrote:
>
>>On August 23, 1999 at 19:29:09, James Robertson wrote:
>>
>>>On August 23, 1999 at 17:26:17, Andrew Dados wrote:
>>>
>>>>On August 23, 1999 at 15:28:10, James Robertson wrote:
>>>>
>>>>>Yesterday on FICS I noticed scrappy (Crafty on P200x4) was playing a human who
>>>>>had already defeated it twice at 5 5 blitz and was mopping up a third victory. I
>>>>>was kind of suspicious, so I started talking to him and asked him to play my
>>>>>program. Overall, it was a completely one-sided affair, but I was convinced that
>>>>>he was not a computer when he missed several tactical things my program saw very
>>>>>quickly.
>>>>>
>>>>>I asked him if he was a GM or IM, and no, he said he was unrated(!). Anyway, I
>>>>>have never seen a human outsmart computers so well in blitz without relying on
>>>>>standard anti-computer play. Nor did he shy away from tactics, and often after
>>>>>complex exchanges he would come out ahead.
>>>>>
>>>>>Is he very unusual, or are there a lot of players like him? He also mentioned he
>>>>>had many rated chess playing friends, and did not think he was near GM strength.
>>>>>
>>>>>James
>>>>
>>>> This is one of his games when he agreed to play with no increment.... 3 0 tc
>>>>and he of course got flagged.
>>>>Note he never moves below 3 sec (as fast as someone relaying moves to chess
>>>>program can go, even most obvious moves), never took more then 7 sec/move.
>>>>Black is typical 'human' time usage. Btw... I'm not accusing him of cheating -
>>>>just bringing some facts to ponder :)
>>>>
>>>> 1. d4 (0:02) d6 (0:01)
>>>> 2. e4 (0:05) Nd7 (0:00)
>>>> 3. Nf3 (0:05) Ngf6 (0:01)
>>>> 4. Nc3 (0:06) e5 (0:01)
>>>> 5. Bc4 (0:03) Be7 (0:01)
>>>> 6. O-O (0:03) c6 (0:01)
>>>> 7. a4 (0:06) h6 (0:01)
>>>> 8. a5 (0:05) Qc7 (0:01)
>>>> 9. Be3 (0:07) O-O (0:01)
>>>> 10. Bb3 (0:05) Re8 (0:02)
>>>> 11. h3 (0:05) Nf8 (0:01)
>>>> 12. Qd3 (0:04) Ng6 (0:04)
>>>> 13. Rfd1 (0:05) Bf8 (0:01)
>>>> 14. d5 (0:05) c5 (0:03)
>>>> 15. Nb5 (0:03) Qe7 (0:06)
>>>> 16. c4 (0:05) a6 (0:02)
>>>> 17. Nc3 (0:05) Bd7 (0:01)
>>>> 18. Na4 (0:05) Bxa4 (0:02)
>>>> 19. Bxa4 (0:04) Red8 (0:00)
>>>> 20. Rdb1 (0:05) Qc7 (0:04)
>>>> 21. b4 (0:05) Nd7 (0:02)
>>>> 22. bxc5 (0:07) Nxc5 (0:04)
>>>> 23. Bxc5 (0:04) dxc5 (0:01)
>>>> 24. Qb3 (0:04) Rab8 (0:03)
>>>> 25. Qc3 (0:04) Bd6 (0:03)
>>>> 26. Rb6 (0:05) Nf4 (0:01)
>>>> 27. Qb2 (0:04) Nd3 (0:02)
>>>> 28. Qd2 (0:04) Nb4 (0:01)
>>>> 29. Nh4 (0:04) Bf8 (0:11)
>>>> 30. Nf5 (0:04) Rd6 (0:07)
>>>> 31. Nxd6 (0:04) Bxd6 (0:02)
>>>> 32. Ra3 (0:03) Rd8 (0:10)
>>>> 33. Rg3 (0:07) Kh7 (0:04)
>>>> 34. Qe2 (0:04) Qe7 (0:08)
>>>> 35. Qg4 (0:04) g6 (0:01)
>>>> 36. Qe2 (0:04) Rb8 (0:02)
>>>> 37. Qd2 (0:04) Bc7 (0:08)
>>>> 38. d6 (0:04) Bxd6 (0:04)
>>>> 39. Qxd6 (0:04) Qxd6 (0:01)
>>>> 40. Rxd6 (0:03) Rf8 (0:02)
>>>> {White ran out of time} 0-1
>>>>
>>>>-Andrew-
>>>
>>>The fishy thing I noticed was of course his time usage.... he never used less
>>>than 4 seconds when I was watching. This seems very unusual, and I would expect
>>>any human to speed up if they played 3 0. Obviously he did not....
>>>
>>>The evidence for him being human was also his time use (when I played it varied
>>>a lot), and that he sometimes made tactical mistakes, even after a lot of
>>>thought.
>>>
>>>He led his king on a surprising king march in his second game with JRCP, and
>>>eventually walked it right into perpetual check. I don't know if that is a
>>>computer or human thing to do. :)
>>>
>>>James
>>
>> Tactical mistakes... I can guess you figured those from differences between
>>your ponder score and move played score in your program. This reminds me games
>>played by my program against BountyHunter on ICC at 0 4 time control: looks like
>>CG makes a lot of those and yet wins buncha games... or maybe our programs are
>>being outsearched badly?
>
>This wasn't a case of being outsearched badly, because in at least one case the
>move JRCP thought he would play really looked like it would win another pawn
>(after JRCP pointed it out, even my brain could see that my program had no
>compensation). He chose a move that was cleary not as good. It seems more likely
>that he used the computer for moves that seemed hard, and his own head for the
>'easier' ones. :)
>
>>(I have yet to see *anyone* titled going 3:0 against
>>Scrappy tactically at blitz).
>
>I wonder what program it was? It seems hard to imagine any program doing that on
>a standard machine....
>
>James
It is easier than you think... crafty tries very hard to avoid draws against
human opponents... too hard to try against computers. IE it will accept a
score well below zero rather than repeat a position, which is sound against
humans at faster time controls (avoid the draw, hope for tactical/positional
mistakes). But against computers, all you have to do is set the contempt
factor high (trying to repeat positions) and you will end up in good positions
every time (using a computer) because your program will be trying to repeat,
scrappy/crafty will try to avoid it, and it will make serious positional
concessions to avoid this.
If it sees it is playing a computer (it knows this on the servers, so long as
the opponent has a (C)) it doesn't do this, it sets the drawscore to a normal
value and plays reasonable chess rather than trying to avoid draws when it
should not...
>
>>- skeptical 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.