[Welcome to Sensei's Library!]

StartingPoints
ReferenceSection
About


Paths
ConnectionGames

Referenced by
ComputerGoProgram...
ArtificialIntelli...
ANoviceTriesToWri...
Arimaa
Havannah

 

Other Games Considered Unprogrammable
   

If you think programming Go is too difficult (with all that ko, seki, counting rules, life and death; even the rules are difficult to get completely correct) to start one's own program, here are some alternatives:

Of course a lot of games exist, for which no good computer opponent exists, perhaps no-one ever tried. Often however there's a strategic component in the game that the human player can handle well, but the computer can not, with current methods.

Table of contents


Similar to Go in the aspects:

  • that the branching factor is high
  • once set, a stone stays on the board & doesn't move

Especially interesting are the following two:

Hex

  • (solved up to 7x7, already very strong programs exist, however less strong than humans on boards larger than 9x9)
  • kde-opponent "six-0.5.0.tar.gz" is very good for example
  • See computer-olympiad-pages for a list of papers on hex

Havannah

("unprogrammable"?!, $1000 for computer that beats human)

havannah is very similar to hex:

Is a strong opponent possible with hex methods? -- ab

Dots and Boxes

ilanpi: I find this game similar to Go in many ways. I believe that it might be more difficult than Go, because there is a parity condition where I could give it inherent instability. It has been completely analyzed by computer up to the 5x5 dots case (16 boxes). The 6x6 case (25 boxes) is unsolved as of 2002. I have played these sizes and have found the play quite similar to 7x7 Go, and therefore similarly much simpler than 9x9 Go. Unfortunately, it seems hard to get a good game on anything larger than a 5x5 dots board, so I don't think much concrete is known about Dots strategy for larger sizes (as compared to go, for example). I have written a web site about Dots: [ext] http://cf.geocities.com/ilanpi/dots.html

  • excellent page!

Twixt

another "connection game"; almost no programs exist currently, so they are also weaker than human players; difficulty..?

[ext] http://www.k2z.org play Twixt online, live (>100 players registered)

[ext] http://www.johannes-schwagereit.de/T1.html the program Twixt T1, with source

Some discussion, for example [ext] here.

In general [ext] here and [ext] here (newsgroups).

Variant: Twixt PP, no links are ever removed, own links may cross other own links; this may be easier to program, with those few and difficult special cases left out, where links had to be removed.


Similar to chess

While the upper two were for the Go players, the following one is for the chess players

Arimaa

  • (a new chess-like game with larger branching factor)
  • there's a prize of $10000 for a computer program that defeats a human player before the year 2020
  • [ext] http://arimaa.com/

Shogi

Anonymous opinion: I've heard this is hard for the computers too.


Other

ktron

ktron: Seems very difficult to me too. Although this is an "action"-game, if you remove the time pressure from the human player you can also look at it as a strategy game, played on a graph, with branching factor 3 and possibly large depth.

Given the condition for winning (being the last player with a legal play), Tron would seem to be well-suited to the kind of analysis so popular on SL.


Zertz

Gerhard: The Zertz game ([ext] http://www.gipf.com) is also difficult for computers since you must look a large number of moves (ususally > 10) ahead. The different winning possibilities and strategy changes during the game can make this a computational difficult problem. I know of only one Zertz-playing program that plays extremely poor.


Links

[ext] http://www.google.de/search?q=computer+olympiad Computer (Game Programming) Olympiad(s)


See also

Artificial Intelligence Programming
List Of Components A Computer Go Program Must Have



This is a copy of the living page "Other Games Considered Unprogrammable" at Sensei's Library.
(OC) 2004 the Authors, published under the OpenContent License V1.0.