<< Back to General Forum   Search

Posts 1 - 12 of 12   
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 04:05:00


Nex
Level 60
Report
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 04:09:54


Nex
Level 60
Report
To be specific:


Look into the line, 1.e4 c5 2.Na3!! Nc6 3.d4 cxd4 4.Bc4 Nf6 5.h3 Nxe4 6.Bxf7+ Kxf7 7.Qh5+ Ke6 8.Qg4+ Kd5 9.c4+ dxc3 10.Be3 and you will find much entertainment
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 08:09:34


Major General Smedley Butler
Level 51
Report
wat
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 08:36:58


Rodtod18
Level 28
Report
I actually did not know that this tactic still worked.
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 10:02:41


szeweningen 
Level 60
Report
That seems very weird. My Stockfish is very reluctant to play 7. ... Ke6 giving an equal position, instead he plays 7. ... Kg8 showing almost -3. And to be precise it did not take more than a few seconds to get that evaluation of Kg8 being superior to Ke6.

EDIT: Ah, yes, the article is from 2001, back when we still had human vs machine matches and thought maybe humans could hold.

Edited 6/13/2015 10:05:44
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 16:47:23


Nex
Level 60
Report
It is saddening that this line is out of favor with modern engines. My Komodo agrees, 7 ... Kg8 is much stronger.


On an amusing side note:

Komodo 9.0 is in a gauntlet tournament vs some other Komodos, Stockfish and Houdini, and currently has an 80% score, as White, after 10. Be3. Komodo agrees with Nemeth, that this is a great gambit for taking down engines! xD
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 17:43:48

Hennns
Level 58
Report
And even though, if the engine correctly evaluate the threat at ply 15., we just have to go back 1 or 2 ply earlier and it is blind again...
-2nd link.

Now obv this was a few years ago (2009), but I don't think whoever wrote that realize how little 15 ply is. 1 ply = 1 half move (ie one side move). So 15 ply is about 7.5 moves deep- that's nothing for a computer! Let a normal computer sit for 30 seconds and it'll surpass that. This is not some amazing flaw in the analysis of computers, he's simply not allowing them to think deep enough. For some comparison, a normal ply by a normal computer nowdays is around 20-30 (depending on the complexity of the position). In this position (move 16) I reached 30+ after analyzing for about 2 min.. Also from after 16. Qh5 it's a forced mate in 11 moves, 16... Qa5+ leads to mate in 10, it's not like that's a bad move :p





As for 1.e4 c5 2.Na3!! Nc6 3.d4 cxd4 4.Bc4 Nf6 5.h3 Nxe4 6.Bxf7+ Kxf7 7.Qh5+ Ke6 8.Qg4+ Kd5 9.c4+ dxc3 10.Be3 moving the King out on the middle of the board is something even the freshest chess players know not to do. Not only does Ke6 serve no purpose, but Kd5 doesn't either- better to just move the King back. 7... g6 is also better (along with 7...Kg8), or 8... Kf7. Having only 80% win rate after Be3 doesn't matter much if your opponent won't enter that line :p Maybe it worked in 2001, but it won't work today.

Edited 6/13/2015 17:45:11
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 19:02:12


Norman 
Level 57
Report
So 15 ply is about 7.5 moves deep- that's nothing for a computer! Let a normal computer sit for 30 seconds and it'll surpass that.
For some comparison, a normal ply by a normal computer nowdays is around 20-30 (depending on the complexity of the position)

Without reading the links, I have some "feeling" how deep computers can see into the future from my WarLight AI. With normal minimax (without Alpha-beta pruning) I can look about 4-5 steps during picking stage in the future (for picks only with different rules than in WarLight) until the time explodes. A depth of 20-30 has to mean that the chess computer is extremely smart in throwing away non promising paths and not that he actually evaluated every outcome with a depth of 20.
How to defeat chess AI: Nemeth's Gambit: 6/13/2015 19:19:30

Hennns
Level 58
Report
Yeah, it only analyses variations it deems "worthy". You know more about programming than me, but the way I understand it it works like a three diagram, and if then one variation is bad- it can throw away the whole branch. (for example, if after 1. f3 e5. 2. g4) the computer finds Qh4# and then determines that g4 is bad- there's no point in analyzing it further. Extreme example-but it shows the point. Additionally it works by choosing candidate moves- which they're surprisingly good at. In many ways they play like humans- only analyzing a few important moves and following variations. And yes, they're extremely smart about analyzing the right variations. If you want to see the actual code, Stockfish is an open source engine- might be worth a look.
How to defeat chess AI: Nemeth's Gambit: 6/14/2015 01:46:43


George Adgdgdgwngo
Level 49
Report
I am a big fan of chess, (I represent my Province) But Why is this is "General" Should be in "Off-Topic"
Yeah I said it...
How to defeat chess AI: Nemeth's Gambit: 6/14/2015 02:35:30


Fleecemaster 
Level 59
Report
I can make it on-topic.

Who is going to design a better AI for Warlight?
Yeah I said it...
How to defeat chess AI: Nemeth's Gambit: 6/14/2015 02:35:57


Fleecemaster 
Level 59
Report
Double Post, and I have no idea how to delete it, so I'm just going to edit it :)

Edited 6/14/2015 02:37:19
Posts 1 - 12 of 12