Table Games in 1016: a hnefatafl AI tournament

Two of my interests, the Old Norse tafl family of games and computer science, combined to create this thread, and the event I’m pushing.

By way of explanation, I’ll start with the tafl games. Played across Northern Europe (and into the British Isles) from about the late 3rd century to the 12th century, attested in several Viking sagas, before being wiped out by chess in the early 1200s. Unfortunately, the saga authors didn’t think to leave us rules, so it seemed the game was lost to history. Fragmentary archeological finds continued through the early 1900s, but then a historian named Harold Murray connected a description of a game called ‘tablut’, from Carl Linnaeus’ diary of his time among the Saami peoples, with the hnefatafl mentioned in the sagas, and we were off to the races. Fast-forward another century, and the tafl games have an extremely small but loyal following, including yours truly. In recent years, a few organizations have crowned world champions, and interest in the game is certainly at its highest point in almost a thousand years.

Much of my fascination with it comes from how it plays. The two sides in a tafl game are asymmetric. The attackers, or the besieging side, start along the edges of the board. Their goal is to prevent the defenders’ king from escaping. The defenders, with half as many pieces as the attackers plus one king, start in the center, and aim to get the king to an edge space or a corner space, depending on the variant. All pieces move like the rook in chess, and captures are made by surrounding a piece on two sides rather than moving into the same space. This website has more information on the rules and history. This one has correspondence play against other humans, if you’re interested in seeing how it plays; if you’d rather beat a bad computer player, you can play against mine.

The other part of my fascination with tafl games comes at them from the computational perspective. Although we’ve had solid rules for one rules variant for a century, there has been literally zero formal study of them beyond one blog post I banged out over a few lunch breaks last year. The upshot is that tafl games are almost certainly harder for computers to play than chess, and very likely to be a lot harder than chess, and, again, have been studied by nobody in the modern age. They are therefore fertile ground, and in the hopes of kickstarting some interest from the theory community (and, somewhat more selfishly, improving the single-player experience and getting more people interested in the game to begin with), I’m running the event in the thread title, more information on which can be found behind the first link in the thread.

If this is a field that interests you as much as it does me, it would be nifty to have your entry, or to have a hand in getting the word spread around. Thanks for reading all this.

(Cleared with Tom ahead of time, in case you were worried.)

I can’t code AI, but I can post pictures of random games I have sitting on my desk!

Hey, I’d post this over at BGG as well. Bound to be one or two coder types over there! :)

Nice set (I don’t think I’ve ever seen a picture of the Viking Game-branded one, although I’ve heard of it), and thanks for the tip! I’d totally forgotten BGG has a tafl section.

(Shameless bump and update.)

I’ve finished a first pass at the external AI mode in OpenTafl, so if you were waiting on the crystallization of that spec, it’s mostly there now.

Another shameless bump and update: I have a prerelease version of OpenTafl’s network play update posted. There’s a server running at intersect.manywords.press, which is the default setting—or at least, there’s a server running until it inevitably crashes. The server lobby UI is unfinished and not everything is tested, but it ought to work at least passably well, if you’d like to try to organize a game (with me, or with someone else).

In ongoing shameless pluggery, OpenTafl has progressed somewhat since then. Network play is just about fully featured, and there’s now a prerelease version with playable variations, so you can explore bits of the move tree. This series of releases is going to wrap up variation stuff (mostly getting them to work with saved games and loaded games), and hopefully add a puzzle mode and a few tafl puzzles.

In AI tournament news, I have three firm entries and three soft entries, which ought to make for an interesting series in December. I have some thoughts now, but they’re probably wrong, so I’ll write about them nearer the tournament once I’ve seen some of the entries.

Quick heads-up if anyone is interested in whipping up an entry: the deadline is in 30 days. I expect the tournament play to happen between Christmas and New Year, and I’ll be covering some of it by stream or liveblog.

Tournament coverage begins with an introduction to J.A.R.L*, the first contender to be submitted. All tournament coverage will be at this page. Tournament play for real will probably start on the day before Christmas Eve.

*I really hope it stands for something clever, but the README doesn’t say.