We recently had a programming competition where each team had to write an A.I. for the board game Broadside. It was perfect because the rules were simple enough that the teams had to concentrate on the basic idea of when and how to attack, and not on handling a number of esoteric rules. At the same time, it had really good gameplay rules so that the smarter A.I. had a strong advantage.

So, any suggestions for games that would be a good one for a programming competition? A perfect game would have:

  • Lack of existing readily available AI's
  • Handles 6+ players
  • Simple easily understood rules (no or few edge cases)
  • Enough depth so that the better AI has a strong advantage
  • 2
    Hmm. Chess and Go have pretty simple rules. But the smarter AI tends to have a strong advantage. I assume you don't want to go down that route. Can you elaborate on your requirements (why are those unsuitable candidates)? Jul 15 '11 at 22:37
  • I don't want to use a game where there are a lot of good AIs out on the web. If it was chess or go it becomes a contest to see who can Google the best open source AI. I also would prefer one that can have 6+ players a board so we can do the play-off quicker. Jul 16 '11 at 1:39
  • Would you prefer games that have little to no randomness, or does that matter? Jul 16 '11 at 18:45
  • Randomness doesn't matter as long as a better designed A.I. will win. In other words writing an A.I. for "Deal or No Deal" would be totally random. But one for Risk would be fine where the dice and cards have a big impact. (Risk is out because the map is too weird for a 9 hour programming contest.) Jul 16 '11 at 20:34
  • 2
    Why was this considered not constructive?
    – Pureferret
    Jan 20 '12 at 23:42

Robo Rally (rules) could be interesting. Amusingly, the game itself is programming a robot, one of several (up to 8) on a game board.

Each player gets 9 random instruction cards per round and chooses 5 to play. Then all the robots' moves are evaluated simultaneously (more or less). Your robot can get pushed around or damaged by other robots as well as by conveyor belts and other board elements. First robot to hit all the checkpoints in order wins.

You should have a few knobs to tweak here to make the challenge easier or harder:

  • Do the players know the board in advance, or does their AI have to deal with it dynamically?
  • The damage rules could be left out for simplicity, or left in for more complexity

It's possible this is too complex, but it's pretty well suited for this kind of contest, I think.

It's also an excellent board game to play in person, especially with other programmers.

  • It's perfect - just the level of complexity wanted, up to 8 players, etc. Unfortunately there are a number of open source versions out there with an A.I. player. Bummer because this would have been perfect :( Jul 16 '11 at 20:51
  • Oh hey, I didn't think of that. Maybe you could tweak the rules somehow to make those ineffective...
    – lilserf
    Jul 16 '11 at 21:19
  • That's a really good suggestion. There are some changes that would make any existing AI useless. Of course, I have to hope I don't create poor game play in that case. But this is worth thinking about. Any suggestions on how to change it? Jul 16 '11 at 21:31
  • Main problem is the complexity of the boards themselves -- getting that info into a form the AI can readily use is more work than the AI itself...
    – Chris Dodd
    Jul 21 '11 at 22:43
  • We used Robo Rally - details at blogs.windwardreports.com/davidt/2011/08/… thank you for an excellent suggestion. Aug 27 '11 at 21:58


Since it's been designed to be hard to implement an AI for, Arima is the ideal choice for a serious (non-recreational) competition. And, if it can beat the best human, it's a ten-thousand dollar prize. Further, the designer WANTS people to try. It also uses chess pieces and a chess board, so graphics are a non-issue, as stock graphics or even standard symbol font characters can be used.

For free rules, and information on the Arimaa Challenge, go to http://arimaa.com/arimaa/

  • 1
    Interesting. My worry here is that there are AIs out there so the winner migh just be the best at google. Otherwise it would be really good. Jul 16 '11 at 20:37
  • Just a matter of comparing the source code to eliminate that...
    – aramis
    Jul 17 '11 at 21:47
  • 1
    @aramis: And what if someone has the same algorithm as an open source version? If you disallow it, you've placed arbitrary restrictions on solutions - and forced everyone to be aware of all the existing AIs in order to not infringe. If you allow it, well, it'll work if and only if everyone's honest - or as long as the winners are all better than all existing AIs.
    – Cascabel
    Jul 21 '11 at 0:37

What about Saboteur?

It is for 2-10 players, very simple rules, the randomness is limited by the "tiles", but allows for quite some AI options.


World Cup Tournament Football

The rules are straightforward and very simple, and success in this game is dependent upon figuring out how your opponents are playing without being able to see which teams your opponents "have". An unskilled AI can tip its hand early in the game in much the same way an unskilled rummy player will reveal their goals to a skilled opponent: they can deduce which teams you've got if you are not careful about how you play your cards.

An additional advantage is that teams can be evenly divided among 6, 8, or 12 AI, depending on how you want to distribute the teams ... you could modify the bonuses given to each team to make teams more or less even so that random draw of teams has less or more impact, or even group teams into specific packs and have each AI play from different positions, similar to a duplicate bridge tournament (maybe even down to distributing cards identically for each run).

Formula Dé

The rules are relatively straightforward, and while the distance you move is largely controlled by dice, there's a good bit of strategy involved with respect to the best path to take, how to approach turns and such. An AI that does not judge turns and relative positions of cars will be significantly handicapped. With the advanced rules, if you allow "players" to determine the number of Wear Points assigned to each component, there should be additional variance between AIs.

  • Formula De has an official Java implementation already.
    – aramis
    Jul 17 '11 at 5:00
  • I was thinking of Formula De too. I've not seen the Java implementation before though.
    – tttppp
    Jul 17 '11 at 8:24
  • boardgamegeek.com/thread/170270/formula-de-online tells you how to get it from formulede.online.free.fr
    – aramis
    Jul 17 '11 at 21:46
  • @aramis, that's not necessarily a bad thing. If an existing implementation isn't open-source, it can be used as a point of comparison or even as an opponent; if it is, then it's basically the same situation as with Arimaa. Jul 18 '11 at 12:54

How about Blockus? It's simple, expandable to any number of players, even to 3d, and has interesting choices on representations of the pieces. It also allows for varying board shapes and sizes in the AI world. And it's pretty to watch running.



It's a card game for 3-6 people. The rules are relatively simple. First you try and predict how many tricks you'll take in a round. And then you play to make it so. If you hit the mark you get points proportional to the number of tricks you made. If you're off in either direction you loose points proportional to the number you're off by.

From what I could gather there is only an IPhone version out there that has AI play.


Also very simple rules and you could even have two leagues: One with movement and one without movement. There are definitely algorithms out there but I don't know how accessible they are...


Based on the comments perhaps this should just stand as "reasons why Diplomacy is not a good suggestion", summarized as:

  • the AI would be too hard to code
  • AI for diplomacy already exists

Original answer:

How about Diplomacy? 6+ players, I've never heard of any AI, rules are deterministic, and room for AI to really achieve and show difference between programmers seems ample.

  • Diplomacy relies too much on negotiations, treaties, and other hard to implement elements, especially given the conditionals in most treaties. It wouldn't end up playing like diplomacy, even tho it might look like it, unless someone had the backing of the Jeopardy bot team...
    – aramis
    Jul 17 '11 at 5:00
  • @aramis good point. I could see the details of how to code up treaties and negotiations as being too difficult in practice. It would be like a multi-dimensional prisonsers' dilemma problem times 1000.
    – Adam Wuerl
    Jul 17 '11 at 12:05
  • 2
    Diplomacy AI does exist. See daide.org.uk/wiki/Main_Page Jul 18 '11 at 0:37
  • lol - hilarious edit Adam
    – Ryan
    Aug 3 '11 at 4:11
  • @Ryan, just trying to stem the legitimate tide of down votes while simultaneously embracing the SE concept of editing answers that are out-of-date.
    – Adam Wuerl
    Aug 3 '11 at 11:35


Ra has simple mechanics, but interesting AI since it's a competitive auction.


Not the answer you're looking for? Browse other questions tagged or ask your own question.