Dr Eric Silverman

UPDATE 6 Nov 2021 — More sample games updated — Dai Dai Shogi, Tai Shogi, Mini Tai Shogi.

UPDATE 25/10/2021 — Some of the sample game GIFs on this page broke, possibly from WordPress again changing the size limit for displayed images.  I’m slowly replacing the sample games for Maka Dai Dai Shogi, Tai Shogi and Mini Tai Shogi with new, smaller GIFs.  These games were generated with much longer thinking times (120 seconds) so the games are also higher quality.

USEFUL TIP: WordPress handles GIFs in a weird way. When you want to see/download the full-size GIFs, click the image and then click the (i) button, and finally click the link to view the full-size GIF in another tab.

---

Since my previous post on our addition of multiple variants of Shogi to the Ai Ai general game-playing software, Stephen and I have been hard at work implementing even more ancient Shogi games.  If you thought the giant 15×15 Dai Shogi was over-the-top, you’ll be amazed what we’ve brought to you this time!

Before I show off what we’ve done, a few points to keep in mind when you download the new Ai Ai version with all these amazing games:

1. We’ve done a huge amount of work in a relatively short period of time, and have done everything we can to catch mistakes and bugs, but some bugs will still be present!  If you find a bug or incorrect piece movement or something, please do point it out to us, and we’ll do our best to resolve the problem.  Also, some rules are still not quite there (repetition rules, for one), so please be patient.
2. The AI is very much a work in progress!  Some of these games are truly gigantic, and as a result getting strong play out of the Shogi engine in Ai Ai is really, really difficult.  When you do play against the AI, be sure to give it plenty of thinking time; at an absolute minimum, give it 30 seconds per move for the 16×16/17×17 games, 1 minute for 19×19, and 2 minutes or more for 25×25.  If you have an old computer with a slow CPU you may need to amp those times up significantly.  If the thinking times are too low, you may find the AI repeats moves a lot as it struggles to find useful continuations.
3. For those games which have a playing community and multiple possible rulesets (Tenjiku Shogi), we have not attempted to resolve any long-standing rules disputes, but instead have opted for being comprehensive.  All the games have options to choose different rules interpretations, piece movements, etc.  Essentially I decided to include a rule proposal or piece movement when that interpretation had been played before, and/or had some historical plausibility, and tried to remain neutral on the long-standing rules disputes in some of these games.  The hope is that players may try all these different variations in Ai Ai, and hopefully use those experiences to choose the best ruleset to play as a community.

With that out of the way, let’s get to the games!  For each one I’ve included some details about the options available to players in Ai Ai.  At the end of the article there is an FAQ section, so please check that for some additional useful tips and tricks for using Ai Ai, and some general questions about our Shogi implementations.

Tenjiku Shogi (16×16)

• 

• 

• 

Tenjiku Shogi (天竺大将棋, or Tenjiku Dai Shogi in Japanese) is a remarkable historical Shogi variant that is one of the most popular of these ancient games among modern players (alongside Chu and Dai Shogi).  Tenjiku Shogi is played on a 16×16 board with each player starting with an army of 78 pieces of 36 different types.  The game stands out not only among Shogi variants, but among Chess-like games in general for its extreme tactical sharpness and super-powered pieces.  The word ‘Tenjiku’ means ‘Indian’ but is more often translated as ‘exotic’, because at the time of Tenjiku’s invention (sometime in the 16th-17th century, most likely), India was seen as a mystical and exotic place, the birthplace of Buddhism and the home of marvellous creatures like tigers and elephants.  Tenjiku certainly lives up to its name, and has some very unique pieces that give the game a distinctive flavour:

1. Fire Demon — for a start, this incredibly powerful piece is very mobile; it can slide unlimited squares in six directions, or take three successive King-like steps in any direction to dodge around obstructions.  But much more notable is its ability to burn opposing pieces — when the Fire Demon ends its move, all adjacent enemy pieces are immediately removed from play.  This means the Fire Demon may kill up to eight pieces in a single turn (one on its destination square, and seven in the surrounding squares).  This ability even functions on the opponent’s turn; any enemy piece that ends its move next to your Fire Demon is immediately burned and removed from play, and that doesn’t count as your turn!  If your opponent lands their Fire Demon next to yours, their Fire Demon is burned, and it doesn’t get to burn any of your pieces before it is removed from the board.  Fire Demons can still be captured by any piece that lands directly on its square.
2. Range-Jumping Generals — Four pieces in Tenjiku, the Great General, Vice General, Rook General, and Bishop General, have the ability to make range-jumping captures, where they may leap over any number of friendly or opposing pieces to land on an enemy piece and capture it.  To keep this amazing ability under control, these pieces are subject to a hierarchy, and may not jump over pieces above them in that hierarchy; this allows players to block these powerful jumps with careful defensive placement.  In the two main Tenjiku rulesets there are two different interpretations of this ability; more on that in the discussion of game options below.
3. More powerful Lion pieces — Tenjiku includes all the pieces present in Chu Shogi including the remarkable Lion, which may move and capture twice in a turn.  In Tenjiku the Lion can promote to an even more powerful piece, the Lion Hawk, which may move as a Lion or a Bishop.  Tenjiku also includes the Free Eagle, a piece that may slide an unlimited number of squares in any direction, or may make a Lion-like double-move but only diagonally.

Combine all these powerful pieces and you get a game that despite its size is remarkably fast-paced and violent, with deadly tactical combinations possible right from the beginning of the game.  Most large Shogi variants have quite long and subtle openings, but in Tenjiku one opening mistake might lead to an embarrassingly early checkmate.  These unusual properties have enabled the game to develop a small but dedicated Western playing community, and as a result numerous resources are available for prospective players, including opening guides.

• 

• 

Rulesets and Options

Tenjiku Shogi has had a lot of attention over the years since it was introduced to the West in the 1980s by George Hodges.  Unfortunately, as with many of these ancient games, some important rules questions remain unresolved.  Modern players have developed two main rulesets that are in use, both of which are available in Ai Ai:

1. Richard’s PBEM Server ruleset: the yearly Tenjiku Shogi Championship tournament is played on Richard’s PBEM Server, and uses this ruleset.  In brief, the main distinguishing features here are: Fire Demons may slide unlimited squares diagonally or vertically; range-jumping generals may capture Kings or Princes while jumping, allowing for some very early smothered mates to occur; and the Heavenly Tetrarchs do not have a vertical sliding move (and in fact use an odd set of movements that seems to be a mistake, so you may want to use the TSA Tetrarch variant instead).  Early checkmates are very common in this ruleset.
2. Wikipedia/Chess Variant Pages ruleset: these rules have been promoted by HG Muller, and produce a somewhat less violent Tenjiku experience than the PBEM ruleset.  In this ruleset: Fire Demons may slide unlimited squares diagonally or horizontally; range-jumping generals may not capture Kings or Princes while jumping, eliminating the early smothered mates; and the Heavenly Tetrarchs have a vertical sliding move, making them a clear upgrade over the Chariot Soldiers from which they promote.

Various spirited arguments have taken place, and continue to take place, over these key rule interpretations.  In Ai Ai we have chosen instead to implement both options, to give players a choice of how they want to play — simply load TenjikuShogi(PBEM).mgl for the PBEM ruleset, or TenjikuShogi(WP).mgl for the Wikipedia/Chess Variant Pages rules.  In the hope of helping the community to resolve these rules discussions, we have also included a customisable version (TenjikuShogi(Custom).mgl), which allows players to customise their ruleset with the following options:

1. Fire Demon move — you may choose whether the Fire Demon slides diagonally and vertically (PBEM version), or diagonally and horizontally (WP version).
2. Range-Jumping Generals — you may choose whether they can capture the Royal pieces (King and Crown Prince) by jumping (PBEM) or not (WP).
3. Heavenly Tetrarchs move — you may choose from four (!) different move options for the Heavenly Tetrarchs:
   1. PBEM move — no vertical slide, may never move to the eight adjacent squares, may capture without moving on adjacent squares, adds a three-square vertical jump (which seems like a mistake)
   2. Wikipedia/Chess Variant Pages move — vertical slide, may never move to the eight adjacent squares, may capture without moving on adjacent squares
   3. TSA move — the original move advocated by The Shogi Association — no vertical slide or vertical jump, may never move to the eight adjacent squares, may capture without moving on adjacent squares
   4. Sho Shogi Zushiki move — vertical slide, may never move to the eight adjacent squares, and may NOT capture without moving on the adjacent squares
4. Lion Hawk move — you may choose two variants of the Lion Hawk:
   1. TSA move — the Lion Hawk does NOT have Lion Power, but instead may move as a Bishop or step twice as a King in any direction, but may only capture once during this two-step ‘area move’
   2. Modern move — the Lion Hawk moves as a Bishop or a Lion, with full Lion Power
5. Free Eagle move — you may choose two variants of the Free Eagle:
   1. TSA move — the Free Eagle may move as a Free King, or may jump two squares orthogonally
   2. Modern move — the Free Eagle may move as a Free King, or may make two successive one-square diagonal steps, which can include making two captures, capturing an adjacent piece and moving back to the starting square (igui capture), or passing a turn by moving to a diagonally adjacent square then back to its starting point.
6. Allow Zone-Passing — you may choose whether or not to allow zone-passing; if zone-passing is allowed, that means that Lion-Power pieces may promote if they make a Lion-style move into the promotion zone and immediately back out again on the same turn.
7. Water Buffalo promotion burn — if Water Buffalo promotion burns are allowed, that means that when a Water Buffalo promotes to Fire Demon, it may immediately burn all adjacent enemy pieces on the move where it promotes.  This is suggested in the Wikipedia/Chess Variant Pages ruleset.

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

That’s a lot of options!  I’m 100% sure that some members of the Tenjiku community would prefer that I adopted only their preferred ruleset/piece movements, but again I have tried to avoid wading into any long-standing rules disputes here.  I opted to go over all Tenjiku-related materials I have and simply include any rules/movements that have some historical backing, or have been played before, or advocated seriously as a proposed change to the rules, so that players may try them all and make an informed choice.  I did decide not to include the original TSA rules for the range-jumping generals, however, as these have been shown to give an easy win to Sente.

Promotion Rules and 50 Move Rule

Note that I have decided to offer all three main Tenjiku variants with the Chu Shogi promotion rules; this is mainly because there seems to be some momentum lately to adopt these rules in the PBEM Server version, and the Japanese Wikipedia rules state the Chu Shogi promotion rules are applied in this game.  If anyone wishes to use modern Shogi promotion rules instead, simply open the appropriate .mgl file in a text editor, and find a line that looks like this:

“promotionRule” : { “rule” : “ChuShogi”, “allowIguiPromotion”:false },

Remove the “rule” : “ChuShogi”, part and you will be using modern Shogi promotion rules.

I also decided not to include any equivalent to the 50-move rule in Chess, mainly because these rules are a modern invention and definitely not part of ancient Shogi, and also because we have no idea what a good move threshold would be in these huge games.  However, if the community does want to use a 50-move rule equivalent, this can be added to any Shogi variant by opening the appropriate .mgl file, finding the section called “endCondition”, and within the square brackets adding another line that looks like this:

{ “condition”:”boredom”, “who”:”all”, “result”:”draw”, “moves”:50 },

Simply change the move number to your preferred option, and that’s it!

In any case, I’m a strong believer that debating game rules without playtesting them is pretty pointless, so I’m hoping that having all these options available will help the various rules debates to eventually get resolved.  With these different rules variations now playable online as well as against the AI, players may thoroughly test them and decide together how they want to play Tenjiku in the modern age.

Tenjiku Sub-Variants

For players who are interested in playing Tenjiku but find the full-sized game intimidating, we’ve also included two smaller modern variants of Tenjiku Shogi:

Nutty Shogi (13×13)

• 

• 

• 

Nutty Shogi was invented by HG Muller in 2015, and is a very enjoyable reduction of Tenjiku Shogi to a more compact 13×13 board.  All rules remain the same, so it’s a straight subset of the full game.  In this implementation players may choose from the options present in the custom Tenjiku Shogi version, in order to match their preferred ruleset for the larger game.   Note that the game seems to have been designed to use the Wikipedia/CVP rules, so I recommend using those options.

Makyou Shogi (12×12)

• 

• 

• 

I’m a fan of Nutty Shogi, but find it mildly inconvenient in that no historical Shogi variants use a 13×13 board, so it’s not easy to play physically even with my collection of Shogi variant pieces and boards.  So I created Makyou Shogi, a reduced version of Tenjiku designed to fit on a 12×12 Chu Shogi board.  Makyou is a Japanese word meaning ‘demon-infested place’, which felt appropriate for this tight board packed with powerful pieces!  Note that the piece mix is a bit different from Nutty Shogi.  Also, this game is still being tested, so the starting array may change over time.

Makyou Shogi allows players to tweak any of the rules/piece options present in the Tenjiku Shogi customisable implementation; having said that, I strongly recommend using the Wikipedia/CVP rules in this game, as the PBEM pieces make the game too tactical with little room for strategy.

Dai Dai Shogi (17×17)

• 

• 

• 

With Dai Dai Shogi, we enter the realm of the truly gigantic Shogi variants.  Players start the game on this huge 17×17 board with 96 pieces each of 64 different types (4 additional types appear only via promotion).  Among those pieces are the ultra-powerful hook-moving pieces — the Tengu and the Hook-Mover — which may make two successive moves as a Bishop or a Rook respectively, making a right-angle turn after the first one (but may only capture once).

Dai Dai Shogi is the smallest of the ancient Shogi games to use a promotion-by-capture rule.  Instead of promoting upon reaching the enemy camp, pieces in Dai Dai Shogi promote immediately after they capture any opposing piece (if the piece has Lion Power, the promotion happens at the end of their complete move).  Promotion by capture is mandatory and cannot be deferred.  Unusually for a Shogi variant, most pieces in Dai Dai Shogi do not promote, including Pawns; this relative lack of promotable pieces together with the promotion-by-capture rule gives the game a quite different feel from the other historical Shogi variants.  The promotion-by-capture rule has a profound impact on tactics, too, as players must now be careful that any capture exchanges don’t leave the opponent a strong promotion at the end of the fight.

The Dai Dai Shogi starting array is packed with pieces and very asymmetric, with many pieces only appearing once rather than in pairs.  The initial position is also arranged quite differently from the other large Shogi variants (with the exception of Tai Shogi, to an extent) — the strongest pieces are at the back of the player’s army, rather than the front. This means opening play is quite slow and subtle, as players try to develop their pieces soundly without leaving any easy captures for the opponent to use for promotion, while also trying to free their most powerful pieces.  The endgame can become pretty hectic, as the hook-moving pieces finally come out to play and their incredible mobility makes them a deadly threat.  All told the game is thoroughly engrossing, deeply strategic and yet filled with rich tactics:

”This is the writer’s personal favourite of the [large Shogi variants], owing to the tremendous variety of pieces, the wealth of strong pieces and weak pieces with strong promotions, and the asymmetrical opening setup…. What often happens in practice is that the entire board opens up into the most complex tactical struggle seen in any of the forms of Shogi.” (R. Wayne Schmittberger, writing in Shogi Magazine in 1981)

Rulesets and Options

Fortunately, unlike Tenjiku Shogi there are relatively few major rules disputes in Dai Dai Shogi, if only because very few people have played it.  As a result our implementation has just two main incarnations: one which includes the troublesome Great Elephant piece in four variations (DaiDaiShogi(complete).mgl); and another which excludes the Great Elephant piece entirely, which apparently is the case in two historical sources (DaiDaiShogi(noGE).mgl).  In both versions the player may choose between two variations of the Furious Fiend’s move, as well.

• 

• 

• 

• 

• 

• 

Alongside the Elephant variants described in the translation notes article, in this final version we also included a variant move sometimes used in Tai Shogi,  which allows the Great Elephant to  move as a Lion Dog or slide up to 5 squares horizontally or diagonally backward.  This move is recommended by the Japanese Chu Shogi Association (Chu Shogi Renmei).

Sub-Variant: Cashew Shogi

• 

• 

• 

Invented by HG Muller in 2015, Cashew Shogi is a reduction of Dai Dai Shogi onto a smaller 13×13 board.  This variant includes most of the characteristic pieces of the larger game, but substantially reduces the number of moves one has to remember.  Despite the size reduction, the AI vs AI test games I’ve tried are often not that much shorter than the full-sized games of Dai Dai Shogi, and sometimes are substantially longer!  I suspect this is mostly due to the size difference, however; the much larger board size in Dai Dai Shogi gives the AI a hard time, so it tends to make more serious blunders during play.  In Cashew Shogi the AI can find stronger moves, and is less likely to blunder its way into a quicker loss.

Our Cashew Shogi implementation allows players to choose their preferred form of the Great Elephant and Furious Fiend, as in Dai Dai Shogi.

Maka Dai Dai Shogi (19×19)

• 

• 

• 

• 

Let’s get my biases out there straight away — Maka Dai Dai Shogi is a very cool game, and I’m overjoyed that it’s now playable in Ai Ai.  This is a truly immense game, played on a 19×19 board with 96 pieces of 50 types in each player’s starting army, and it uses a promotion-by-capture rule like Dai Dai Shogi (more on this below).  What makes Maka Dai Dai Shogi stand out is the presence of several unique mechanics and pieces that give this enormous game the feel of a giant mythical battle; this is no coincidence, as the game is clearly heavily influenced by Buddhist mythology:

1. The All-Powerful Emperor — unlike the other Shogi games, in Maka Dai Dai the King can promote!  If your King captures an enemy piece, he may promote to Emperor, perhaps the most powerful piece in any Chess-like game.  The Emperor can instantly jump to any square on the board, including squares occupied by enemy pieces, but it may not enter protected squares or capture a protected enemy piece.  Checkmating the Emperor seems impossible, but the hook-moving pieces can make it happen.
2. The Deva/Dark Spirit/Teaching King/Buddhist Spirit — At the start of the game the Kings are flanked by a Deva and a Buddhist Spirit, two incredibly weak pieces with very awkward asymmetric 1-square moves.  But when a Deva captures an enemy piece, it becomes a Teaching King, a super-powered piece that may move as a Lion Dog or a Free King; likewise, the Dark Spirit becomes a Buddhist Spirit, which moves as a Lion or a Free King.  More intriguingly, these pieces are contagious — that means that if an enemy piece captures your Deva or your Teaching King, it immediately becomes a Teaching King, or if it captures your Dark Spirit or Buddhist Spirit it immediately becomes a Buddhist Spirit.  Consequently these pieces are very difficult to eliminate from the board permanently, unless they are taken by another Teaching King/Buddhist Spirit or by a royal piece (King, Emperor or Prince, who ignore the contagious aspect and simply promote normally).  Some Shogi historians believe the Buddhist Spirit and Teaching King may represent the Buddha and the Lotus Sutra, respectively.
3. Promotion by Capture — as in Dai Dai Shogi, pieces in Maka Dai Dai Shogi promote when they capture something, so there are no promotion zones (or, alternatively, we can consider the entire board the promotion zone).  However, there are two variants of this rule: in one version, suggested by Wikipedia and the Chess Variant Pages, the capturer may choose whether to promote or not, unless the captured piece is promoted, in which case they must promote; in the other, suggested by the TSA rules pamphlet and Japanese Wikipedia, pieces must promote when they capture, as in Dai Dai Shogi.  In either case, Maka Dai Dai moves away from the Dai Dai Shogi model and allows nearly all pieces to promote, more like the other historical variants.  Many short-range pieces promote to ‘Free’ versions of themselves, which move in the same directions but extend all one-step moves into unlimited sliding moves.

These elements combined give Maka Dai Dai Shogi the strategic depth and nuance of the other large games, but punctuated with moments of extreme dynamism: the Emperor can teleport around the board savaging his enemies; the two powerful, mystical spirits stalk the board and never stop hounding one player or the other; and even the weakest pieces can promote into powerful board-spanning threats.  I honestly can’t recommend this game enough, it’s huge and crazy but very very interesting and enjoyable to play, and just packed full of cool pieces.  Here are my picks for the top ten coolest pieces in Maka Dai Dai Shogi, from strongest (Emperor) to weakest (Free Bear):

• 

• 

• 

• 

• 

• 

• 

• 

• 

• 

Fortunately, it seems I’m not the only one who developed a bit of an obsession with this game.  In the historical documents there is a long chant given that helps players to remember the opening setup, which suggests the game was popular enough that players developed these kind of mnemonics to help them get the game started more quickly.  A research group in Osaka, led by Professor Tomoyuki Takami, has been studying this game for years, generating a lot of lively debate about its origins, influences and rules; Professor Takami has also invested a lot of effort in promoting the game.  I have a fervent hope that one day these pockets of interest in Maka Dai Dai Shogi will blossom into a full-blown playing community, and this Ai Ai implementation is my attempt to help that process along.

As a side note, Prof Takami’s group has developed a set of rules for what they believe to be an earlier form of Maka Dai Dai Shogi, which they call Maka Dai Shogi.  This game also seems very interesting, but we haven’t included this in Ai Ai because it would involve programming numerous variant rules and pieces, and the rules are not stable and seem to change fairly frequently as Prof Takami finds new evidence in the historical documents.  At some point we may include it, when I have time to reach out to Prof Takami and get a comprehensive set of rules.

Rulesets and Options

Maka Dai Dai Shogi has relatively few options to worry about when starting a game:

1. Promotion Rule — players may choose whether promotion on a capture is compulsory, or whether it is generally optional but becomes compulsory when the captured piece is promoted.
2. Furious Fiend move — as in the other large games featuring this piece, players may choose the old TSA Furious Fiend move (Lion + 3-step slide move) or the more current interpretation (Lion + Lion Dog).

That’s it!  Note that the old TSA rules included a variant Teaching King move which was nonsensical, as it had the piece moving like a Free King or using a 3-step slide, which completely overlap.  The one historical Japanese source I have available clearly indicates a Lion Dog move plus Free King move, and all other sources I can find suggest this move, so I’ve opted not to include the TSA move here.

Sub-Variant: Macadamia Shogi (13×13)

• 

• 

• 

• 

Invented by — you guessed it — HG Muller in 2015, this 13×13 reduction of Maka Dai Dai Shogi retains all the most distinctive pieces in the game and removes most of the weaker ones.  The consequence is a quite intense game, where a nerve-wracking opening phase tends to develop into a rather intricate middlegame as the players try to outfox one another with their remaining nimble power-pieces.  I’ve enjoyed my time with this game quite a bit, and definitely recommend trying it.

Our implementation of Macadamia Shogi allows players to select the same options as the main game (promotion rule and Furious Fiend move).

Sub-Variant: Hishigata Shogi (19×19)

• 

• 

• 

Hishigata Shogi was invented by Sean Humby in 2005, and attempts to slim down Maka Dai Dai Shogi in a different way.  In this game the full-size 19×19 board is retained, but the initial position excludes nearly all the weaker pieces, and the King is placed much closer to the centre of the board and surrounded by his army.  The result is an extremely fast-moving game, as the power-pieces swoop dangerously around the board, picking away at the enemy King’s dense pack of protectors until they can find a fatal weakness.

Personally I substantially prefer the original game to this variant; Maka Dai Dai Shogi shines partly because the deadly power-pieces are embedded in vast armies of weak-yet-still-important short-range pieces, balancing out their ridiculous power somewhat thanks to the levelling effect (referred to in my first post on our Ai Ai implementations).  Hishigata abandons much of that, and I feel the result is a bit less nuanced.  However, I can imagine some players will vehemently disagree with me and will enjoy the frenetic tactical explosion this variant offers, so of course I include it here for everyone to try for themselves!

Upon starting a game of Hishigata Shogi players may choose the promotion rule to use, but there is no choice of Furious Fiend moves as that piece doesn’t appear in this game.

Tai Shogi (25×25)

• 

• 

• 

• 

• 

Tai Shogi is a truly immense game, and is the second-largest Chess-like game ever created (36×36 Taikyoku Shogi being the largest).  On its vast board of 625 squares, players battle it out with dense armies of 177 pieces each, with 93 different piece types present in the initial position.  Tai Shogi essentially combines the pieces present in Dai Dai Shogi and Maka Dai Dai Shogi, and adds an additional nine new types of pieces.

Until now I haven’t had the opportunity to play a full game of Tai Shogi, so I’d always seen it as phenomenally impressive, but probably too big to be truly playable.  However, since implementing it in Ai Ai I’ve played with it quite a bit, and I have to say I was wrong; Tai Shogi may be huge, but it’s definitely playable!  Not only that, the sheer size and scale makes the game feel quite epic, and I’ve enjoyed my time with it so far.  For those of you out there who are familiar with Dai Dai Shogi and Maka Dai Dai Shogi, please do give this game a try; the learning curve will be small, since there are only nine new pieces, and you’ll be able to develop some basic strategies based on your experiences with those games.

Tai Shogi has a couple unique properties of note:

1. No Kings on the board — instead of starting with a King, players in Tai Shogi start with both an Emperor (!) and a Crown Prince on the board, both of which must be captured to win.  Since the Emperor can jump instantly to anywhere on the board, that means it effectively serves as a potential extra protector for every friendly piece, adding an additional tactical wrinkle to any battles taking place.  Games also tend to run long (2,000 moves is a typical length for a game), since the Emperor is far too powerful to get checkmated in the early stages when few pieces are developed and available to attack it.
2. Full complement of hook-movers — Tai Shogi players have at their disposal the Hook Mover (double Rook), Capricorn (double Bishop), Tengu (double Bishop + single orthogonal step) and Peacock (double Bishop, but only forward).  With six hook-moving pieces on each side of the board, long-distance captures are always a threat, especially later in the game as the board empties out a bit.  However, players must use these pieces cautiously — some will demote on capture, and they can’t be spent carelessly as otherwise checkmating the opposing Emperor will become very difficult!

Playing Tai Shogi is quite an experience, and it’s most definitely the most wargame-like of the many Shogi variants I’ve played so far.  The starting armies are so huge that the board tends to have multiple local skirmishes happening at the same time, and the advancing Pawn lines backed by swarms of generals and mythical beasts reminds one of the clash of phalanxes of ancient soldiers.  The game clearly has scope for incredibly varied strategies, so I’m looking forward to playing more so I can begin to appreciate its subtleties.

Rulesets and Options

Tai Shogi has two main rulesets in use, which substantially change the gameplay:

1. Japanese Wikipedia rules — These rules essentially combine Dai Dai Shogi and Maka Dai Dai Shogi in their entirety.  Most of the pieces from both games are included, and they promote as they do in their parent games, meaning that nearly all pieces in the game promote.  Promotions are compulsory upon capturing an enemy piece, as in Dai Dai Shogi.
2. TSA rules — In the TSA version of the game, promotions are limited to about 1/3 of the pieces in the starting array, similar to Dai Dai Shogi (Pawns don’t promote here, either).  Promotions are also compulsory on capture, like Dai Dai Shogi and the Japanese Wikipedia rules.

The additional 60 or so promotions available in (1) make the game feel substantially more hectic than the TSA rules, and since I don’t have the historical sources available to decide on way or the other, I decided to include both rulesets in separate files so that players may choose for themselves.  Ruleset (1) is in TaiShogi(JWP).mgl, and ruleset (2) is in TaiShogi(TSA).mgl.  Try them both and see what you think!

Note that English Wikipedia has yet another set of promotions, but the entire page appears to be in a state of flux and so I don’t consider those suggested promotions reliable enough to include as a third option.  Similarly, the German Chu Shogi Association appears to suggest yet another variant where Crown Princes may promote to Emperors, but I can’t find much support for this idea elsewhere.

Players may also choose between variant piece movements for several pieces:

1. Great Elephant — this has the same four move options as in Dai Dai Shogi
2. Furious Fiend — this has the same two variants as in other games including this piece
3. Fierce Eagle — players may choose between the TSA move given for this piece and the Japanese Wikipedia move
4. Great Dragon — players may choose between the TSA move and Japanese Wikipedia move

Given the sheer size of this game, I wouldn’t worry too much about which moves you choose; the changes between each option are minor and will have little to no impact on the course of a full game on this massive board, particularly when there are numerous other pieces with substantially more power flying around.

Sub-Variant: Mini Tai Shogi (15×15)

• 

• 

• 

• 

This work-in-progress variant is my own creation, and is an attempt to shrink down the Tai Shogi experience onto a smaller 15×15 board.  The most exciting pieces from Tai Shogi are all present, and players start with Emperors on the board and multiple hook-movers, plus Poisonous Snakes and Old Kites that can promote to hook-moving pieces in the endgame.  My test games thus far have lasted anywhere between 400 and 1,000 moves, so still a long haul but substantially shorter than Tai Shogi.

At the start of the game, players may choose variant moves for the Great Dragon, Furious Fiend and Great Elephant.  Players can also decide to replace the Emperors with Princes, if they want a less mind-bending endgame; however, I strongly recommend using Emperors to get the full Tai Shogi experience!

As with Tai Shogi, there are two versions of Mini Tai Shogi — one following the promotions set out in Japanese Wikipedia, the other following the TSA rules.  The Japanese Wikipedia version has nearly all pieces promoting, while the TSA version has more limited promotions available.

Mini Tai Shogi is definitely a work-in-progress, but so far I’ve found it enjoyable to play.  Tai Shogi is such a huge game that reducing it down to 15×15 means eliminating a huge number of pieces from the board; over time I will experiment with different piece choices and arrangements for Mini Tai, and may even produce a second smaller variant with a very different set of pieces drawn from Tai Shogi.

Emperor Endgame Trainer

Maka Dai Dai Shogi and Tai Shogi add a new endgame twist over the other ancient Shogi games — the mighty Emperor, who can jump instantly to any unprotected square on the board.  At first this piece is very confusing to deal with, and checkmating it can seem almost impossible.  To make the transition to Emperor endgames a bit easier, and to open up a way to have some quick fun with all these crazy Shogi pieces, we added an Emperor Endgame Trainer to Ai Ai that helps you learn how to cope with Emperor endgames.

When you load up the trainer (EmperorEndgameTrainer.mgl), you’ll be given a menu where you may choose five different pieces for Sente to use to try and checkmate Gote, who starts with an Emperor and two Gold Generals.  If you have Ai Ai play as the Emperor side, you can experiment with different piece combinations and learn how you can checkmate in these wild new endgames.

• 

• 

• 

Rulesets and Options

Upon loading the trainer, you will be able to choose five pieces for Sente to use, from the entire current list of available Shogi pieces in Ai Ai.  However, to keep things applicable to the actual games that use the Emperor, only the pieces available in Maka Dai Dai and Tai Shogi can promote, and they use the promotion-by-capture rule.  At the start you may also select whether to have promotion-by-capture as compulsory (like Tai Shogi) or optional (like Maka Dai Dai Shogi).

You may also choose whether one or both sides have a King or an Emperor, if you just would like to gain some experience with some of the Shogi variant pieces in more normal endgame situations.  The board size is configurable too.

Note that if you leave piece selections on ‘Random’, you may occasionally get starting positions where the enemy Emperor/King can be captured immediately; for example, selecting a Bishop hook-mover like a Capricorn, Tengu or Peacock as Sente Piece #2 will cause this.  If that happens, you can click the Game menu and select Game Parameters to choose different pieces and try again.

Shogi-Inspired Chess Variants

For those die-hard Chess players who miss the classic FIDE pieces and those wonderful FIDE Pawns, don’t fret!  I’ve also put together some highly experimental Chess variants that try to capture the large Shogi feel in a distinctly Chess-flavoured package.  These games use Chess pieces and popular fairy pieces, but adopt large Shogi conventions like very dense starting setups, promotions for many other pieces besides Pawns, and eschewing castling in favour of starting the Kings in strong fortresses right from the beginning.

All three games are really just intended for fun, so please don’t take them too seriously; having said that, I have playtested them quite a bit and they’re playable and not obviously broken.  The starting positions have been tweaked and tuned to avoid any obviously weak squares for either side to target in the opening, the piece mixes have been adjusted following test games, and promotions have been swapped around in each game to try and encourage the use of certain pieces.

Rider Romp (10×10)

• 

• 

This game is sort of a bridge between 10×10 Chess variants like Grand Chess and Opulent Chess and the large Shogi experience.  The opening setup is denser than most 10×10 Chess variants, and the pieces are more powerful and many are capable of promotion.  The four middle Pawns have been replaced with Stewards, variant Pawns that can move and attack in four directions, which are quite robust defenders; in early testing, a line of ten normal Pawns proved to be somewhat feeble resistance against the onslaught of all these power-pieces.

The name comes from the game’s theme of promoting one’s pieces to stronger forms that include additional ranging moves.  Some of the promoted pieces are ‘rider’ pieces, a term originating from fairy Chess which means they extend a basic move by allowing the piece to perform it repeatedly.  Nightriders are a classic example — they are Knights that can perform repeated Knight leaps in the same direction in one turn.

Rider Romp starts out fairly tactical, with long-range attacks flying everywhere; just mind the Steward Pawns in the middle four files.  Once the board empties a bit promotion really becomes a focus.  These games use promotion zones like the large Shogi games, so players will have to try to advance their own pieces to achieve strong promotions, while trying to guard their own huge promotion zone against incursions from enemy pieces with high-impact promoted forms.  In my tests a game played out to the bitter end takes about 150-250 moves (using Shogi-style move counting, not Chess style!).

Neutronium Chess (12×12)

• 

• 

Now things get a bit crazier.  Neutronium Chess gets its name from the ridiculously dense material that makes up neutron stars, because this game is absolutely packed to the brim with pieces at the start of the game, and there are 33 distinct piece types available for use.  The piece lineup includes various strong leaping pieces, some with additional orthogonal or diagonal step options to add flexibility, and powerful sliding pieces as well.

• 

• 

• 

• 

• 

My favourite pieces in this game are the Griffon and Manticore, which are bent slider pieces — the Griffon takes one step as a Bishop then can slide unlimited squares outward as a Rook, while the Manticore takes one step as a Rook and then can slide outward as a Bishop.  This gives them some very surprising attack options, but also gives them a fundamental asymmetry; Griffons and Manticores can’t retreat the same way they advanced, so they must be deployed with finesse.   This game (and Dai Chess) also contains other bent pieces, including the Crooked Rook and Crooked Bishop that move in zig-zag patterns, and the Ship (invented by Jean-Louis Cazaux here), a vertical-only Griffon.

As with Rider Romp, many pieces promote in Neutronium Chess, so watch out for strong pieces getting even stronger in the endgame.  In play this game has an unusual feel, with the phases of the game proceeding in a Shogi-like progression, but the high-powered pieces create complex, long-range tactical shots more reminiscent of Chess.  I’m still not sure it’s ‘good’, as such, but I do have a lot of fun playing it.  A typical game lasts anywhere from 200-400 moves; the sample game shown above is on the shorter side, finishing at 261 moves.

Dai Chess (15×15)

• 

• 

Dai Chess takes the next logical step and increases the board size to a massive 15×15.  This game uses the exact same starting squares for pieces as Dai Shogi, but with all different pieces.  All Pawns start the game protected, so there are no immediately obvious weak points to target in the opening.   Players start play with 65 pieces each, and there are a total of 46 piece types available for use.  Given the larger board size I wanted to try for a more strategic feel in this game, so some of the more powerful pieces from Rider Romp and Neutronium (like the Griffon and Manticore) now only appear via promotion, calming the early stages of the game somewhat.  One aspect I’m still working in is the placement of the colourbound pieces, which is troublesome due to the board having odd dimensions rather than even, so keep an eye on that in future revisions.

Dai Chess really amps up the large Shogi feel of this series of variants, not just by using the Dai Shogi position but also by adding more promotions.  Nearly every piece promotes, and in keeping with what we expect from Chess, all promotions are an improvement in strength; I’ve avoided any demotions in the style of Dai Dai/Tai Shogi.  This game makes use of a lot of leapers, so good tactical vision is necessary to avoid early loss of strong pieces.  In the endgame, sliding pieces become dominant as they can promote much more easily, which again is reflective of the large Shogi inspiration.

A typical game of Dai Chess lasts about 400 moves, sometimes substantially more; the sample game above lasted 495 moves.  Ai Ai can actually play this game reasonably competently at 15-30s thinking time, but it can end up being a bit greedy in the opening; if you want a greater challenge, trying playing a cautious, positional opening for Ai Ai before turning the AI on.  Be careful though — for human players all these weird leaping pieces can make it hard for us to spot good moves, but the AI doesn’t have that problem!

Expanded Chess/Get Bent/Symmetric Sissa

• 

• 

• 

• 

• 

• 

If you’d like to experiment with those interesting bent-slider pieces some more, you can also check out my variant called Get Bent, also added to Ai Ai in this release, which includes a whole mess of them on a 10×10 board; it’s a very fast-paced tactical game.  I mainly made it to get familiar with these odd pieces, so it’s just for fun.  Pawns promote to extremely powerful Griffon/Manticore compound pieces, so watch out for that!

We have also included Expanded Chess, a nice variant by Daniel Zacharias which has Griffons and Manticores as well, and Symmetric Sissa, a showcase for the multipath Sissa piece, which on each move must take the same number of steps as a Rook and a Bishop (in either order).

None of these three variants have anything to do with Shogi, but since they’re in the new release I felt it was worth highlighting them anyway!

---

Next Moves

Sorry for the insanely long post, but as you can see, Stephen and I have done a whole lot of work on these games over the last few months.  As I mentioned at the beginning of this article, bugs will very likely remain in one or several of these games, but we’ve at least reached a stage where the games all function and we’ve incorporated every rule/piece option that seems plausible.

From here, we hope you Shogi fans will give these games a try, let us know how you find them, and maybe organise a tournament or two.  If/when any bugs crop up, please let us know of course!

I’m sure some folks would prefer there were less options to deal with in each game, but honestly I can’t see any other way to go; the alternative would be to decide unilaterally how these games should be played, but the information we have on the historical rules is very fragmented, so I don’t feel qualified to do that.  I also don’t want to take sides on the online debates regarding these rules, which have been going for far longer than I’ve known about these games.  In any case, my hope is that these options can be trimmed over time as the community tests them out and reaches a consensus on what pieces/rules are most enjoyable.  Similarly, as I get my hands on the historical sources I’ll revise things here and there as needed.

After this release we’re taking a little break from Shogi programming, but at some point down the line we do hope to bring in the legendary Taikyoku Shogi as well.  That will take a lot of research on my part, given there seem to be major differences in the rules in each of the three main historical sources, but I look forward to learning more about that massive game along the way.

In the meantime, go play some ancient Shogi, and have fun!

---

Frequently Asked Questions

“How can I choose a Shogi variant to play?”

On the File menu in Ai Ai, click ‘Choose Game’.  A new screen will open with a file browser on the left and a preview display on the right.  Open the ChessFamily folder on the left, then open the Shogi sub-folder.  Within that you will modern 9×9 Shogi, and two sub-folders, one for Historical variants and another for Modern variants.  Pick your favourite, double-click it and Ai Ai will load it up!

Alternatively, hit Ctrl+Shift+F (Cmd+Shift+F on Mac) and start typing the name of the game, and you should be able to load it that way.

“How can I play online?”

Instead of running ‘ai ai.jar’, run ‘online.jar’ which is included in your Ai Ai folder.  Once it starts up, click the button on the left to create an account, then click the link in the subsequent confirmation email (check your spam folder if you don’t see it).  Once your account is confirmed you can login and start playing!  Ai Ai supports correspondence-style online play.

“I’m playing Tenjiku Shogi, how do I use igui capture with my Heavenly Tetrarchs?”

Click and drag the Tetrarch on top of the adjacent piece you want to eat, and when you let go of the mouse button that piece will be igui captured.

“I can’t visualise these crazy moves!  Can Ai Ai show me where my pieces can go?”

It sure can!  There are three main options for this:

1. On the ‘Settings’ menu, turn on ‘Show available moves’.  This will highlight any pieces on your side that have available moves on your turn, and if you click and hold on that piece, Ai Ai will show you what squares it can legally reach.
2. Also on the ‘Settings’ menu, turn on ‘Highlight moves from position’.  When this option is on and you click and hold on one of your pieces, Ai Ai will dim the rest of the board and show you where your piece can move using arrows and will highlight the possible destination squares.
3. On the ‘Game’ menu, choose ‘Available moves…’ and a window will open up that will show you a visual depiction of every possible move in the current position; you can choose your move by clicking through the different options and clicking OK.  Alternatively, simply press Ctrl+M (Cmd+M on Mac) to open that display.

There are many useful display options on the Settings menu, so do make use of them to help you get started with these games.

“How can I change the AI settings?”

On the AI menu, you can choose whether to play an AI vs AI game, human vs human, or human vs AI.  You can also use the buttons on the bottom left of the screen to select these options.  Click on ‘Set thinking time (s)…’ on the AI menu to determine how long the AI may think about each move.

“How can I see all the piece types present in each game?”

On the Reports menu, click the ‘List Pieces (Chess family only) option.  After a few seconds, Ai Ai will generate an HTML table showing all pieces present in the current game you have loaded, including their names, piece graphics, abbreviations in the notation, and Ai Ai’s estimate of their material value.

“Why are the piece abbreviations in the move notation different from the previous convention?”

In Ai Ai we need to define promoted and unpromoted forms of pieces separately.  A Lion that appears in the starting position is defined as a separate entity from a Lion that appears when another piece promotes.  In most previous implementations, promoted pieces appear in notation as ‘+[piece promoted from]’, so when a Kirin promotes in Chu Shogi, for example, the resulting Lion is notated as ‘+Kr’.  However, piece promotions often vary across this set of games, so if we wanted to maintain that we’d need to make lots of duplicate pieces to account for these various promotions.

So instead, we opted to abbreviate a piece that appears after promotion as ‘[promoted piece abbreviation]+’, since this allows for consistency across all the games, and no additional duplication of pieces in the code.  So our friend the Lion appears as ‘Ln’ when he starts on the board, and as ‘Ln+’ when he appears by promotion, regardless of what piece he promotes from.  In these historical variants this works fine, because promoted pieces never unpromote, so we don’t particularly have to care what piece they promoted from anyway.

This also means that if new large Shogi variants arise in the future, we can easily add in games with new promotions without needing to add duplicate pieces.  That in turn makes it possible for enthusiasts to make new variants with different promotions without asking Stephen or I for help writing additional code.

However, the situation changes in games with drops, like Tori Shogi, modern Shogi, or Wa Shogi.  In these games, pieces in hand will demote to their previous form, so we *do* need to know what pieces the promoted pieces used to be.  So, in these games pieces maintain the standard notation.

“How do I make those cool animated GIFs of my games?”

At the end of your game, click the Game menu and choose the ‘Review…’ option.  Once the game review screen opens, click the icon that looks like a strip of film to create an animated GIF.

“How can I tinker with the AI and try to improve it?”

On the Settings menu, choose ‘Select Role…’ and give yourself the ‘Superuser’ role.  When you next start Ai Ai, you will have a greatly expanded array of menus and options, many of which will allow you to perform various analyses on games, generate heuristics, test different AI methods and settings, etc.  There’s a lot to play with, Ai Ai has many powerful features for game analysis 🙂

“I have a great idea for a modern variant of these games!  Can you make it for me?”

If you take a look at the .mgl files that define each game, you can see that creating a new variant is pretty straightforward, so long as it can be constructed out of components (pieces and rules) that Stephen has already implemented.  Give that a try first, and if you struggle to get it working, then I can help you to fix it.

“I don’t like the default options you chose for some of these games, and I’m tired of changing them every time.  Can I change them permanently?”

Absolutely.  Just open up the appropriate .mgl file for that game in a text editor, and find the section at the end called “parameters” — this section defines the options that appear when you load the game.  Each option as a default setting, and if you change that to one of the other options listed in the “optionNames” line of that parameter, you will change that default setting permanently.

Just make sure that you save the changed .mgl file with a different name, and keep a backup copy somewhere; otherwise if you copy a new Ai Ai update into the same folder that file will be overwritten by the Ai Ai update.

“Why does the AI struggle with these large Shogi games?”

Ai Ai is a general game-playing framework, and excels at using variants of Monte Carlo Tree Search.  Chess-like games don’t suit these kinds of algorithms very well, so Stephen has added a general Chess-playing engine to make them playable.  However, because of his engine’s amazing flexibility and generality, each specific Chess or Shogi variant has to be optimised separately to get these best out of the engine, and this takes a lot of time and effort.  At this stage we are mainly concerned with getting the games functional, so we haven’t yet engaged in any game-specific optimisation (that will be my job in the months to come).

The main issue though is simply that these games are huge!  Every position has a large number of possible moves to search, and the AI is simply not able to search as deeply as it can in smaller games.  Extending thinking times does help, but even getting 4-5 moves deep requires far, far longer than in smaller games.  Eventually I hope to experiment with using other AI methods for these games, but this will take a lot of time.

So, for now at least, please be patient and give the AI as much time to think as you can stand 🙂

“Hey Eric, why didn’t you include [my preferred rule]?  It’s clearly superior because….”

I completely understand that some modern players would prefer I chose one set of definitive rules, but the problem here is there’s no consensus on what those definitive rules should actually be, and some of these arguments have been going on far longer than I’ve been involved in the Shogi variant world.  That being the case, I’ve decided to opt for a preservation approach, where any rules proposals that are plausible given the historical information we have on these games are kept alive in these implementations and given as an option.

My hope is that over time, the Shogi community will try these various options and decide on a ruleset for each game that we can consider the modern standard.  At that point, I would be happy to revise these implementations to have one ‘standard’ version for each game, and preserve old options in a secondary, customisable implementation, for posterity.

“Hey Eric, why did you include stuff nobody plays anymore, like Hodges’ moves for the Lion Hawk and Free Eagle?”

If a move or rule has been played extensively and is plausible in some way, then I’ve tried to include it, regardless of its popularity.  I initially wasn’t going to include Hodges’ moves, given the modern Tenjiku community seems to have rejected them, but then I read this old correspondence of his on that very issue:

“I well remember my and John Fairbairn’s discussions with our good friend Maruo all those years ago when we were discussing the more exotic variants in his house.  He was absolutely adamant that [the Lion Hawk] does NOT have Lion Power.  There are several points to make.  Firstly, his interpretation is in our mind in no doubt whatsoever.  Secondly, translated text … says words to the effect “…moves like a Lion…”, now this is simply a short way of describing a two-step mover in all directions.  Lion power is something quite different.”

He went on to explain why appeals to logic (“why would the Lion promote to a weaker piece?”) and symmetry don’t work, because plenty of ancient Shogi games include demotion and substantial amounts of asymmetry.

I found this argument reasonably compelling, and given that George was a scholar with connections to Shogi history researchers in Japan and had direct access to the historical sources, we can reasonably assume his opinions on these matters were well informed.  Therefore, I had to consider those moves plausible, and decided to include them as an option.

“Hey Eric, why didn’t you include my Shogi variant?”

We might do that later, but for now the focus is on the historical variants.  Modern variants, particularly modern variants that are already playable elsewhere, are of much lower priority because they don’t need our help to be preserved and made playable.  Also, speaking for myself I prefer to work on variants I have personal experience with, and I am deeply obsessed with large Shogi and much less so with smaller modern variants 🙂