Human killer engine - a cafè-monster

[JBNielsen]

Has anyone made a "Human killer engine"?

I mean an engine that not always plays the best move because it will:
1) Have open and complicated positions
2) go for positions where it can perform a surprising combination
3) set a trap for the opponent.
4) more things like that

It is an idea I had more than 20 years ago.
It is probably not easy to make, but it could be fun and challenging.

I consider to modify/copy my engine Dabbaba to such a cafè-monster.
With its 2100-elo it will never be a top-engine anyway

Of course such an engine will be easier for a good player to defeat.
But for a weaker player it may often seems even stronger than programs with higher elo.
Because he often looses short games due to the complicated positions and the higher frequence of traps and combinations.

But the weaker player will also have a chance to win now and then.

And not least; it will be more fun to play with.
It will be a complete new feeling, that you have to consider the engine made a bad move by purpose - because it tried to take advantage of the errors that humans make...

[Vinvin]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

[carldaman]

Has anyone made a "Human killer engine"?

I mean an engine that not always plays the best move because it will:
1) Have open and complicated positions
2) go for positions where it can perform a surprising combination
3) set a trap for the opponent.
4) more things like that

It is an idea I had more than 20 years ago.
It is probably not easy to make, but it could be fun and challenging.

I consider to modify/copy my engine Dabbaba to such a cafè-monster.
With its 2100-elo it will never be a top-engine anyway

Of course such an engine will be easier for a good player to defeat.
But for a weaker player it may often seems even stronger than programs with higher elo.
Because he often looses short games due to the complicated positions and the higher frequence of traps and combinations.

But the weaker player will also have a chance to win now and then.

And not least; it will be more fun to play with.
It will be a complete new feeling, that you have to consider the engine made a bad move by purpose - because it tried to take advantage of the errors that humans make...

First of all, I'm glad people are posting about topics that I myself was going to touch upon, independently

We had a recent discussion in this same forum about engines with interesting/unique styles, and yet another one about which engines can be set to play at a (credible) weaker human-like level.

(see links below)

http://www.talkchess.com/forum/viewtopi ... 04&t=47102

http://www.talkchess.com/forum/viewtopi ... 70&t=47140

I think Thinker Active is a great example of this concept, however it is monstrously strong despite its swindle tendencies. Other very strong engines that have a Swindle feature are Hiarcs and Crafty (I think). Another swindle-capable engine, albeit somewhat weaker, is Glass. Hiarcs and LittleGoliath can be set to search for combinations more closely. The tactical mode in Houdini 3 may do a bit of the latter, too.

I'm not sure if any of the weaker engines can actually be set to turn the swindle function on, but the recently released TrappyBeowulf I think does this by default. Zchess/Pharaon and Amyan can actually be set to change their style if playing a human. They are strong engines, however.

Neurosis, a weaker engine of about 2100-2200 strength, has a tweakable config file that can modify the way the engine plays to a human's liking.

Another way to control the engine's play in order to get it to take more risks is by changing its contempt setting. The Zappa Dissident Aggressor personality, recently discussed as well, uses this approach coupled with a very high King Safety and tweaked piece values so the engine will sacrifice material left and right, often unsoundly, but with great effect not only vs strong humans but even vs other engines(!). I've also tweaked Little Goliath to get it to play crazy & exciting chess and it does that, but at a weaker level than Zappa DA, but still strong for a non-master human player.



I do hope you can create such features/personalities for Dabbaba. To me, this should be a legitimate primary goal of (future) computer chess, to get engines to play as much as possible as humans, (including mistakes!). I doubt that most top developers are interested in following this path, as they are mostly focused on gaining pure strength, as measured in rating points. It is great to see a programmer actually raising this question!

best regards,
CL

[mclane]

Has anyone made a "Human killer engine"?

I mean an engine that not always plays the best move because it will:
1) Have open and complicated positions
2) go for positions where it can perform a surprising combination
3) set a trap for the opponent.
4) more things like that

It is an idea I had more than 20 years ago.
It is probably not easy to make, but it could be fun and challenging.

I consider to modify/copy my engine Dabbaba to such a cafè-monster.
With its 2100-elo it will never be a top-engine anyway

Of course such an engine will be easier for a good player to defeat.
But for a weaker player it may often seems even stronger than programs with higher elo.
Because he often looses short games due to the complicated positions and the higher frequence of traps and combinations.

But the weaker player will also have a chance to win now and then.

And not least; it will be more fun to play with.
It will be a complete new feeling, that you have to consider the engine made a bad move by purpose - because it tried to take advantage of the errors that humans make...

Chess System Tal comes to my mind.
Gambit-Tiger too.

The idea was to run into complex situations on the board.
to create trouble.

where the opponent makes mistakes.

[JBNielsen]

In 1998 : http://www.rebel.nl/rebel10c.htm#ANTI-GM

It is a bit like that I want.

But the focus should be more on 1800 elo players than grandmasters.
And much more focus on tactical play and entertaining games than elo performance.



Let me illustrate it more precisely with two of my games from the autumn in 1973 - almost 40 years ago!


GAME 1.

Here are the opening moves 1.e4,e5 2.Nf3,Nc6 3.d4,exd4

[d]r1bqkbnr/pppp1ppp/2n5/8/3pP3/5N2/PPP2PPP/RNBQKB1R w KQkq - 0 4

I continued with 4.c3,dxc 5.Bc4
A 0.43 advantage to black, but the game will probably have action!
This is how the human-killer engine should play.

Later in the game I got a lost position with 2 pawns down, though.

[d]4r3/3q3r/3p2pk/p1pP1pp1/1pR1n3/1P3N1P/P3RPK1/2Q5 w - - 6 40

Rybka4 says white has 4 almost equal good moves.

For me one of the moves was clearly better.
I played 40.Rcxe4(!!) just before the time control.

The game continued 40.-,Rxe4?? 41.Nxg5!! (41.-Rxe2?? 42.Ne6+!! with mate in 3 moves)
The position is even now, and I eventually won the game.

This was hard for black to see, as white had two surprising moves with the knight.
So the human-killer engine should play 40.Rcxe4(!!) for the same reason as I did.



GAME 2.

1.Nf3 Nf6 2.g3 g6 3.Bg2 Bg7 4.O-O O-O 5.d3 c5 6.e4 d6 7.Nc3 a6
8.Rb1 Nc6 9.Be3 Rb8 10.Qd2 Be6 11.Ng5 Ng4 12.Nxe6 fxe6 13.Bh3
Nce5 14.Kg2 Nf3 15.Qe2 h5 16.Bxg4 hxg4 17.Rh1 d5 18.Nd1 d4 19.Bd2
g5 20.h3 gxh3+ 21.Rxh3 g4 22.Rh1

[d]1r1q1rk1/1p2p1b1/p3p3/2p5/3pP1p1/3P1nP1/PPPBQPK1/1R1N3R b - - 1 22

Here I saw the combination that ends the game.
So I played the original move -,Kf7.
I did not expect Rybka4 would like this move.
It soon showed me, that there were 12 better moves...
But it liked it better and better with more calculation...
And at depth 13 it is Rybkas best move!!
That surprised me!
So perhaps it is not an original move, but a deep move...

Rybka4 surely did not prefer -,Kf7 because white COULD fail...
But the human-killer engine should play -,Kf7 because of this possibility!

22.-,Kf7! 23.b3
White will obviously play his knight to b2.
I decided to wait with my trap, so I did not look too eager...
I would like to disquise my intensions..
Good to do something on the queenside and then return to the kingside...
To prevent my opponent from focusing too much on the kingside.

23.-,b5 24.Nb2

[d]1r1q1r2/4pkb1/p3p3/1pp5/3pP1p1/1P1P1nP1/PNPBQPK1/1R5R b - - 1 24

Now I played my trap.
24.-,Rh8
Played a little fast, so it looked a bit as a moved-too-fast error.
Black would normally prefer to play this move with Rb8/Qd8 reversed.

Now white apparently believed he could get control of the h-file (for a while, at least).

25.Rxh8?? Qxh8

White, who later became a 2300+ player, resigned here!

If 26.Rh1, Qxh1+!! with mate by Rb8!


- - - - -


Especially this second game shows how we humans try to fool each other.
(I assume Magnus Carlsen does that with a number of small details!?)
The 23.-,b5 queenside move and moving fast with 24.-,Rh8 was only done because the opponent was a human.
Such things are probably too hard to put into a program.


But the other moves illustrates what a human-killer engine should be capable of!

[Aser Huerga]

But the focus should be more on 1800 elo players than grandmasters.
And much more focus on tactical play and entertaining games than elo performance.

David Kittinger programs fits exactly with your requirements: http://chessprogramming.wikispaces.com/David+Kittinger

[JBNielsen]

But the focus should be more on 1800 elo players than grandmasters.
And much more focus on tactical play and entertaining games than elo performance.

David Kittinger programs fits exactly with your requirements: http://chessprogramming.wikispaces.com/David+Kittinger

Can you explain a bit more about what Kittinger has done?

Will it fx play Rcxe4 as I have shown in another post here?

[Aser Huerga]

But the focus should be more on 1800 elo players than grandmasters.
And much more focus on tactical play and entertaining games than elo performance.

David Kittinger programs fits exactly with your requirements: http://chessprogramming.wikispaces.com/David+Kittinger

Can you explain a bit more about what Kittinger has done?

Will it fx play Rcxe4 as I have shown in another post here?

David Kittinger was the programmer of the best Novag microcomputers, you can find a lot of information about Novag on the www, it's human-like playing style, and specially its good results against 1800-2000ELO players

[JBNielsen]

But the focus should be more on 1800 elo players than grandmasters.
And much more focus on tactical play and entertaining games than elo performance.

David Kittinger programs fits exactly with your requirements: http://chessprogramming.wikispaces.com/David+Kittinger

Can you explain a bit more about what Kittinger has done?

Will it fx play Rcxe4 as I have shown in another post here?

David Kittinger was the programmer of the best Novag microcomputers, you can find a lot of information about Novag on the www, it's human-like playing style, and specially its good results against 1800-2000ELO players

I had both Constellation 3.6(?) and Super Constellation - they were good and had a nice playing style!

But they were not directly searching for traps so they would play like this:

[d]8/8/1P2P3/8/6p1/2r2kP1/7P/1R4K1 b - - bm Re3; id "Jens Bæk Nielsen.39";

1.-,Re3(!!) just seems to threat the e-pawn....
2.b7??, Re1+! with stalemate.


[d]r2q1rk1/ppp1ppbp/2np1np1/8/2PPP3/2NBBQ1P/PP3PP1/R3K2R b KQ - bm Nd7; id "Jens Bæk Nielsen.40";
1.-,Nd7(!!) makes a threat on d4.
2.Ne2?? covers the pawn, but then 2.-,Ne5! gives black a winning position.


[d]N3Q3/3P2pk/5p2/8/2P1p1PK/1P1rP2P/5P2/2n4b b - - bm Nd2; id "Jens Bæk Nielsen.41";
1.-,Ne2(!!) is the only chance.
If fx 2.d8Q?? black mates with 2.-,g5+, 3.Kh5,Nf4+! 4.exf,Rxh3+ mate.

[Father]

JBNielsen,

An engine without opening books,and parameters where knights has an stronger value than rooks, could produce a dangerous moster.