4PC Bots? | From PhoenixZero, To TeamMachina, To TeamTitan, Arnold Schwartzenegger, and more
Before we get started, thanks to @Mugiwara for showing me how to integrate quotes and suggesting the usage of Pixlr, a great software for image editing and, more importantly, creating thumbnails! Really helpful stuff.
The Preface (Introduction)
You don't need a computer to tell you that 4-Player Chess (4PC) is more than just twice as complex as its 2-player counterpart. The game of 2-player chess has been around for centuries, and it's been decades since us mortals could defeat the best of what computer engines can do. Stemming back to Kasparov's 1997 defeat at the hands of IBM's Deep Blue, man is no match for machine in one-on-one chess. So the question arises, how do engines perform in 4-player?
— @grable (Former 4PC Admin), January 26, 2021
If you've ever played 4pc or variants with bots before, such as the Futers or Randos, you know that they aren't particularly good at the game. The Randos make random moves (duh!) and the Futers capture free pieces. Obviously, the Futer is the stronger of the two types of bots, but even still, it is unable to analyze the position to a depth of multiple moves ahead and therefore cannot find the absolute best move at all times. It also does not know opening theory whatsoever, so instead of playing standard ideas, it might make a random knight or pawn move and then bring out a rook.
Incidentally, someone comes up with the idea of an engine for 4 player chess. And if you search up "4 Player Chess Engine", you will get a few results on the first page. A couple YouTube videos here and there, maybe a Reddit post: this content is either educational/entertaining content or a bunch of forums that don't actually have usable AI software, rather just question forums created by a user asking about whether or not such an engine even exists.
Because teams, at its core, is really a 1 vs 1 programmatically speaking (albeit, with 4 colors), an engine can be hard-coded from scratch, trained, and tested, just like a normal engine built for regular chess. I'm not going to get into the details and specifics of the machine learning and computer science behind this because no one wants to hear it. If you are interested, however, I'd recommend you to a YouTube series because I'm not experienced enough for that.
The engines that we see today were created singlehandedly—that is, each engine was constructed, trained, and tested by one person, but they are constantly undergoing revisions and further training through complex data analysis and experimentation. Here is a quick list of some bots and their creators:
PhoenixZero - @omatamix
TeamMachina - @kooser6
TeamTitan (new, still in testing) - @obryanlouis
TeamTerminator (by far the strongest of the engines—for reference it calculates up to depth 12, which is insane for a 4 player game with such a large board) - @Tony
Many additional engines were works in progress and scheduled to come out but never did. One major example would be the Ziva Zero engine, a collaborative attempt at making an all-rounder which was meant to encapsulate teams, FFA, and even 4p giveaway. However, attempts at programming an FFA engine have been made in the past, with little success. It simply becomes exponentially more difficult to develop an engine which must also consider the various factors that could affect gameplay—most importantly, luck. After all, a game of FFA doesn't end until a certain player offsets the balance that is formed and maintained in the 3 player stage. That's why @Radon is not 3300!!! But that's a story for another day.
Some of its Ziva's intended programmers were @GTSWPM150 (now inactive), @kooser6, and @omatamix. For more information on engines, feel free to check out these developers' profiles (but again, some people like GTS are inactive).
What makes these engines interesting is that, while there are many of them, some of them aren't as strong as you might expect, but some of them are destructive machines (that is to say, the "Stockfishes of Teams", but no one calls them that because that would be weird). The sheer variety of strengths and weaknesses can be attributed to the difference in programmers' skill levels, knowledge, and techniques, but still, can we please acknowledge their efforts to not only entertain us and give us insight into how games should be played, but also making the 4pc experience just so much better? And considering these beasts were crafted only around 2 years after the introduction of 4 player chess—impressive, yet freakish.
Behind the Scenes - Q&A With Devs
I asked some of the developers mentioned in the previous sections about:
- The backstory behind their motivation and reasoning to develop an engine for 4pc
- How they began experimenting and programming their machines
- What their expectations for the engines were
- How the bots turned out and whether or not they met the set expectations
...and other important or extra details that they may have wanted to express. Here are some of their responses:
Interview with @Tony, developer of TeamTerminator
TeamTerminator is a chess engine that can play standard 4PC teams chess, and a few 4PC teams variants such as capture the king and 960. It does not play FFA. This engine is just a hobby project of mine. I'm not a 4PC player myself, but I think 4PC teams is an interesting variant.
TeamTerminator is a traditional alpha-beta searching engine, written in C. It doesn't use a neural network. It uses many well known features/techniques of traditional chess engines like material evaluation, mobility, hash tables, null move, late move reductions, etc. However, it's nowhere near as advanced or as optimized as top engines like Stockfish are for 2-player chess.
Before TeamTerminator, I developed a normal 2-player chess engine, written from scratch, which I started over 15 years ago. My 2-player chess engine started out extremely weak (less than 2000 rating) but eventually reached around 2400 rating strength, based on chess engine rating lists.
TeamTerminator development started in March 2019, as a fork of my 2-player chess engine. The first version to play online was from October 2019. I made further improvements to its playing strength during 2020 through 2022. Multiprocessor support was added in February 2021. I haven't really done any work on it during 2023, so far. It reached a rating slightly over 3000 on Chess.com, in 4PC teams.
When the current version plays against the Oct-2019 version, it wins over 95% of the games, so it has improved a lot.
— @Tony (TeamTerminator)
What I find funny is when he says this:
However, it is still not unbeatable in games with human players. I think it's very likely that some other engine or engines will surpass TeamTerminator in 4PC teams strength in the future.
— @Tony (TeamTerminator)
Fun fact: it took Tony more than 10,000 lines of code to program the Terminator! That's a lot of work, but it really shows why the Terminator is just so much stronger than the other engines and even some of the strongest players in the world!
Interview with @omatamix, developer of PhoenixZero
PhoenixZero only has 110 thousand self-play games (compared to AlphaZero's 40 million). With 4 player chess being so complex, it might be a while before PhoenixZero becomes unbeatable. In the future I plan on building a computer which can train the AI’s network faster. I am currently working on making a MuZero implementation of PhoenixZero.
— @omatamix (PhoenixZero)
Some Additional Notes
In the next section, we'll be going through 5 different games played by these engines. Unfortunately, I can't use the diagram editor to display 4pc games, so I linked the games and will be going through key moments via screenshots of positions instead. Each engine has its own collection of a vast majority of games which you can check out by going through the Variants server archive.
I will be walking through the moves with you, so try not to go too far ahead. This is important because there will be some "puzzles" later. I encourage you to try and solve these puzzles before either viewing the solution (click and drag over the spoilers, which are the "empty" gaps between sentences) or watching the game (which can sometimes contain the solution). There are some pretty instructional moments and hopefully the ideas posed in the game can help you improve your own skill.
Move notation is exactly the same as it is in regular chess, except that there are 6 extra files and ranks. I've included the in-line board notation—the letters on the edges of the squares—if you are having trouble following along with the moves, and you can also use the notation/coordinate helper—which highlights the coordinates of square you are hovering over—by going to the game and pressing "K" on your keyboard (make sure you have keyboard shortcuts enabled in settings).
Remember that the engine in each game plays both sides of the team, which is what I mean when I state that there are "2" or "3" players. In reality there are 4 colors, but two of them are played by the engine (and if it's an engine vs engine game then the other two colors are played by the other engine).
By the way, there are going to be 3 Terminator games because they're just too good. The sheer dominative nature that was once possessed by Arnold himself must be withheld, maintained, and developed by his successors. Whether it be dynamics or brute force, the Terminator always gets what it wants. GG.
The Games
I will be honest, this section is the longest by far and makes up the bulk of the post. If you're not really interested in game analysis and instead would rather focus on the lore, or if you'd rather go over the games yourselves, feel free to skip over this section. That being said, it's still very content-rich, with loads of technical tactics and stunning sacrifices, so I would recommend going through the analyses regardless. Maybe save it for later, though.
All 5 games are linked in this miniature "table of contents" (if you can even call it that), and each game is linked separately in the header of the corresponding match that I analyze.
Game 1
Game 2
Game 3
Game 4
Game 5
Game 1 - PhoenixZero vs SeadraCheeseChess & Legend-007
This was a game played between @SeadraCheeseChess, one of my old friends—we even joked about being siblings (long story), but sadly she left after the merge—and @Legend-007, a strong teams player and a friend of mine that still plays today.
The opening was pretty standard.
Phoenix chooses to play the 1. c7, 2. c9 setup followed by a swift queen slide four squares out of the back rank for blue, while for green it chooses to bring the queen out even faster, on move 2 instead of move 3, leaving the pawn on m6 alone so as to not commit the queenside knight to the l4 square. Pretty orthodox, so much so that I'd consider it the "default" opening for 4pc. The human opponents, meanwhile, go for a similar setup, but deviate from the standard ideas on with 3. Nd3 on the board. As mentioned earlier, if you are going to play this setup, your queen knight is basically forced to go to d3, but Seadra does it a bit earlier than expected. This early Nd3 allows her to respond with tempo on Phoenix's blue queen as outlined below. However, it doesn't really do anything.
Phoenix reacts and aptly slides the queen to i10, threatening to checkmate Seadra on i2, so Legend plays Qg13 so he can capture the l8 pawn with check in case Phoenix takes on i2. So far, so good. The next few moves are just developing moves, with RY creating threats/attacks and BG responding to them—lots of queen maneuvering. Fast forward to move 9, and we can see a knight sacrifice on l10, forking the rook in the corner and the bishop—a common idea to open the diagonal and allow the queen to sneak in with mating threats. Most of the time, it isn't really a "sacrifice", rather just a spark to create attacking ideas.
However, rather than moving the rook, Phoenix attacks the red knight...and then proceeds to hang a queen on the very next move. Perhaps it was a calculation error? Or maybe omatamix was trolling with the move generator. Either way, it's safe to say BG lost this game pretty quickly.
They even thought it was a mouse slip!
(Actually, the engine relays its moves to the server by itself. So it wasn't a slip, probably just a calculation error in the early stages of its life. If you were wondering, the bot "talks" because omatamix manually writes the messages lol
Nevertheless, I wanted to highlight this short move-order tactic:
At first it may look like Seadra just hung—or sacrificed, whatever you want to call it—a queen, but upon further inspection you can tell that Legend is going to give a check to Phoenix's green king, allowing Seadra to infiltrate with the queen and end the game faster. And when you're up so much material, you don't really have to worry about losing a rook for a knight.
Anyway, the game ends shortly after, with a swift double queen "in your face" checkmate.
I'm sure the king likes having multiple queens in his face. 💀🤣🤷♂️👍
Game 2 - TeamMachina vs MisterWish & anurag_b
Of course, bots don't always lose games. Even if they're not as strong as you might expect, they are still perfectly capable of pulling off the craziest swindles.
Take this game Machina played against one of the strongest teams players as of now, @MisterWish, and @anurag_b, another great teams player. They're both also friends of mine—it just so happens that within a community so small everyone is friends with everyone else. As incredibly skilled as they are however, (spoiler) even they miss mate-in-one sometimes 
The opening played in this game is strikingly similar in some ways and significantly different in others when compared to the Phoenix game.
Machina immediately brings out the queens, though does it differently as Yellow, playing instead the queen's bishop pawn rather than the king pawn as is standardly played. Meanwhile, anurag opts for the c7 c9 setup (like Phoenix did in the last game), while Wish promptly develops the bishop with potential threats on the f13 square. To prevent this, Machina moves the red queen to f8, defending the pawn from any possible attacks.
The game continues normally, but for some reason Machina slides the yellow king to h14 on the next move. I could not explain the reason behind that move to you except maybe because it was trying to sidestep any check threats from the blue knight?
We soon reach this critical point in the game. anurag plays Ne10, attacking Machina's red queen. Machina then plays d11, attacking the knight, and Wish plays l10, attacking Machina's yellow queen. Such is the nature of teams—tactical and very dynamic. And the tactical attack on the yellow queen allows anurag to later play Ng11, threatening a fork on f13 whilst the green pawn is still attacking the yellow queen. Machina chooses to save the queen, but is then faced with this position:
Do you see the idea? Bonus points if you can find a move that "saves" (because it is still losing for RY) the game. Answer: The idea is Nxf13#, because the blue bishop guards the green queen, and Yellow is in double check. Qxc7+, Qd9, or Qf11 prevents these checkmates—the former is a check that interrupts blue from delivering the checkmate and the latter two block the connection between the bishop and the queen.
Regardless, anurag takes on f13, which forks the king and the rook and leads to a massive exchange of pieces that results in RY being down material:
Fast forward a few moves and the following position is reached with BG up two rooks and a pawn for a bishop:
Now, there is a tactic here that looks like it wins Machina's queen. Can you see it? No answer for this one 
, let me know your thoughts in the comments!
And I'm going to leave off this game here, because red has a swindle defense that actually ended up winning Machina the game, which I don't want to spoil for you guys. Do you see it? (Hint hint, totally doesn't have to do with anything I said in the paragraph before the analysis)
Take a Break...
...because ooh boy, are you in for a ride. Go grab a coffee or make yourself a sandwich. Arnold does not mess around. The Terminator does not mess around.
Game 3 - TeamTerminator vs icystun & E1s6
This is where it gets exciting! The Terminator engine's true offensive ruthlessness, defensive prowess, calculation depth—all of those are displayed in the next three games. This game is played by "the Termy" against @icystun—possibly the greatest 4pc player of all time—and @E1s6, a very strong teams player (who is also working on creating a teams engine at the time of posting this blog—an aspiring developer!).
A Quick Side Note:
You may have already noticed that the starting position is different. This is known as the "Old Setup" or "Old Standard", because this was the standard position before the 4pc/variants merge. One of the major benefits of this position is that there is a lot—I mean a LOT—of opening variety, at least when compared to the current standard position. Of course there are still downsides, one being that (according to Termy itself, this is the worst possible starting position out of 16 possible choices of king and queen swaps, with a beginning evaluation of approximately -4—disadvantage for BG), and there are definitely much better starting positions, but this is the one we ended up with, which makes this one of the most contentious topics in the history of 4pc. Nonetheless, I'll save the rant for later, but if you do want a full, more in depth history lesson, here's @ChessMasterGS's blog on the State of Variants. Let's get back to the game!
Already you can see the major difference in opening choice. Termy goes for 1. h4, which was a popular line back then, with the standard idea being to play f3/Nf3 and g4 sometime in the future. This opening allows for quick development and attacking ideas and basically forces a bishop trade if green were to ever play l8. Elsa goes for the standard c7 approach, and Termy then plays Ni12 as yellow. icy opens with Nl9. Fast forwarding a couple moves and we can see that Termy has not played g4, but delays it to move 7 instead.
When you bring your queen out this early and leave your diagonals open to attack (which is what h4 does), it's not a good idea to further open your king to checks and allowing the opponents to gain tempo and threaten all sorts of tactics. That's why people either play g4 as soon as h4 and Nf3 are on the board, or they wait for the time to come. And as you can see on move 7, icy just pushed his queen up, directly into the line of fire, allowing Termy to safely play g4 with tempo on the queen.
Fun fact: did you know that the Terminator was also programmed to be able to use arrows?
Now I'm not sure why, but Elsa decides to play Ra4 after Termy's g4. Maybe it was meant to be a prophylactic move, prematurely defending b5 with the idea to move the bishop later, but anyhow, Termy suddenly strikes with Nxd11. icy blocks the attack on his queen with l7, and Termy moves the red queen to g9. The point of this move is to pin the knight to the b9 and b8 squares, as if it moves RY can deliver a similar "in your face" checkmate. Now that icy blocked in his own queen, he doesn't have any way to counterattack Termy in case it tries to attack Elsa. If you noticed, a move ago, icy's queen was still open on the diagonal and could sacrifice itself on g2 to give a check had Termy tried to checkmate Elsa (of course, this only happens if the knight moves). You can see this detailed in the screenshot below:
To not fall into either Nxb10+ or Nxc9+, both of which allow major attacking possibilities for Termy (Nxc9+ wins a clean rook by a discovered attack), Elsa slides the king over to a7. However, this allows Termy to play Ne9, attack the blue queen, and while it also puts its own yellow queen under attack, the red queen defends the yellow queen. At worst, it would be a knight for rook trade. icy tries to do something and plays Nj5. I personally think that Nj5 was played too early, because Red can simply take the knight and not have to worry about being attacked while they attack blue, which is exactly what happened in the game. I instead would've probably played Bm7 or Nk7 and waited to play Nj5 after the trades occurred, so at least Green can somehow get some form of counterplay. But anyway, going back to the game, icy plays Nj5 immediately, and Termy snap-takes, while Elsa is still undergoing some major problems, with the two queens staring down at her king and the knight attacking her queen:
This is around the point in the game where Termy converted its advantage into a winning middlegame, a tactical win of material, and eventually a checkmate.
Elsa takes the yellow queen with 11 seconds on the clock, and Termy captures the blue queen with check. icy recaptures the knight on j5, but rather than taking the rook, Termy takes the c9 knight with the red queen first, and only then proceeds to capture the rook on the next move.
Unfortunately for BG, the rook was the only piece guarding the back rank. With it being eliminated, Termy easily sneaks in and takes almost all of Elsa's pieces. The game goes on for only a few more moves. icy tries to save his partner's position, but only manages to prolong the game for a couple more moves before Elsa inevitably gets checkmated.
A remarkable game! This was just one of many matches and rematches played between these 3 players. An engine has managed to defeat one of the best players in the world! But unlike Deep Blue vs Kasparov, the Terminator won almost every single game it has played against icy. This goes to show the true strength of artificial intelligence in chess.
Game 4 - TeamTerminator vs TeamTitan
Engines don't only play against humans, however. Here's a game showing how they fare against each other. Which one do you think is going to win?
Well, I think it's safe to say that Titan got destroyed in this game.
We're back in the standard setup, so the openings become boring again.
Skipping to the first critical moment, Termy begins a checkmating attack with Qf10 and Nd4, threatening Qxc7+ and Nxb5+ or Nc6+, which are double checks and force the king to move. Titan sees the idea and plays Bb7, blocking any Qxc7 check ideas, but downright hangs Nxb5+. Termy immediately takes advantage of this and plays Qf5, which at first looks strange—as if it serves no purpose—but by its next move we can clearly see what the idea is.
If Titan had a brain, it would probably begin regretting its decision by now. What's crazy is that, because it's an engine, it does not have a brain, and therefore does not have emotion or feel remorse! How cool is that?
Titan plays Qd9, right after "realizing" that there is a subtly lurking but dangerous threat; Termy plays Qb9, and the game is over.
That's right.
This position is dead lost for BG.
Titan tries desperately to save the position with Qc10, but it's over. Termy sacrifices the knight with check, allowing it to play Qxc10 and give a discovered check to Titan with the red bishop before blue can recapture. And if you give Termy a full queen advantage...well, you know what it's capable of.
Titan begins sacrificing all its pieces to try and prolong the game, but by this point, it's over. You can't escape from the Terminator. I mean, it even gives up its queen in the most obnoxious way possible, just to try and escape checkmate:
Would I be wrong in saying Arnold is pretty flashy? I'm sure he would appreciate the compliment. So to continue his legacy, the Terminator sacks the queen for a forced checkmate in two moves:
A short game, grotesque and appalling as it may be, but an explosive one nonetheless. Clearly, the Terminator lives for the jawdropping attacking sacrifices like the one you saw in this game, right? But it's not just a tactical genius. It also plays perfect positional chess, which we will observe later.
This game clearly shows how incredibly powerful Termy's calculation and evalution was, but that's not to say Titan didn't play well either. Of course, a one move blunder led to Titan's inevitable devastation, but it wasn't obvious why that move lost the game so quickly. And it was able to see ALL of the moves that prolonged the game for as long as possible, much like those annoying puzzles in Puzzle Rush that ruin your run because you forgot a random rook could block the back rank checkmate and premoved something else. So its calculation is on point as well...with the exception of those rare circumstances where it just "blanks out". In any case, I hope that game was exciting!
Game 5 - TeamTerminator vs icystun & MalburySoapPinkSoap
The finale. Can't end a blog without a banger game.
So icy is back at it again, this time with @MalburySoapPinkSoap, a renowned OG player that everyone knows...well, most people know him. The random noobs don't.
And let's just say that icy gets his revenge in this game. A legendary feat that is yet to have been accomplished by anyone else. Redemption arc, here we come?
We're in the Old Setup once more, because what better way to end off a dynamic battle than with the OG 2018 explosive standard setup that we once used to have?
Oh yes, this game features a SICILIAN DEFENSE!!! FYI: we call it that because from Blue's perspective it looks like the Sicilian and we were too lazy to come up with a better name
The "main" line after Blue plays d6 is Nd3 from Red, which is exactly what Soap plays. The idea is to defend the f2 pawn while at the same time having plans in the future to play Nf4 and either Nd5 or Ne6, threatening checks on Blue. Had Termy instead opened with 1. c6, then Nd3 isn't as useful since the d5 square is defended by the pawn on c6:
So why not just play c6? And that is good question. The drawback of playing c6 is that Blue is unable to play Qb6 and develop the queen, because Yellow can simply attack it with Bh12. On the contrary, had Blue played d6 instead, they would be able to play Qc6—because the pawn isn't on that square, it's on d6—rather than retreating. That being said, neither of the two moves are "better" then the other. It's just personal preference.
But I digress. Termy plays 1. d6, and upon Soap's Nd3, responds with 2. c4, attacking the knight but also guarding the d5 square. Termy plays the same setup on the other side, and icy reacts the same way, with Nk12. Instead of playing l11, which makes the position symmetric, Termy deviates and plays Qm9. Soap continues with the Nf4 idea and Termy develops the queen to c5, allowing Ne6 with tempo, but it has another plan in mind. Or, I should say, in its artificial mind.
icy copies Soap and plays Ni11, but instead of reacting to that threat, Termy rolls up with the queen to h4. Soap goes Ne6 as planned, and Termy then plays Qf5. icy plays Qf12, threatening to play Nc7+ and pick up the blue queen, so Termy defends it with Qe4, also threatening a possible checkmate on Soap's king (Qxh3+ Qxg2#, but not immediately—icy can stop this by playing Qf3 blocking Termy right before it attempts to checkmate with the green queen).
At this point, there aren't any tactics yet. Nc7+ doesn't really work, but Soap sacrifices the knight anyway, and icy trades queens with Termy after it captures the knight. If you haven't noticed by now, icy is down to just over 5 seconds on the clock. This man plays the next 45 moves with just 5 seconds and 15 seconds of delay per move, which is not good, especially when you start with 2 minutes and go down to 5 seconds within the first five moves. But watch just how ruthlessly this guy starts to play.
Fast forward a couple moves, and both teams start developing pieces, looking for attacks on each other. icy is now down to 3 seconds, but Termy blunders! On move 9, it plays Nxd11, thinking that the pawn defends the knight, but icy sacrifices THE ROOO—just kidding im not a gotham fanboy (but the rook sac still happened)—OOOK! And Termy can't recapture because Soap plays Qh4, attacking the blue rook in the corner!
So RY have won the knight back and are playing for a win! In fact, Termy incorrectly plays Nl6 and captures on k4, allowing Soap to recapture and icy to play Nk7, threatening a fork on m6. And now things aren't looking so good for BG. But icy is down to one second left on the clock, and the game still goes on for another 40 moves! But the fireworks have just begun...
It's move 12, and Soap "sacrifices" the queen on m4. It's a "sacrifice" in quotes because icy then plays Nxm6+, which is just a clean win of a rook. And this might look dead lost for Termy, but the game isn't over until you've won, especially when you're up against Arnold himself.
Termy begins to unleash its own counterattacks with Bxj2. I guess icy either got scared of something happening to the red queen or he's a calculative genius—most likely the latter—but either way, he sacrifices the knight on l8. After Termy takes the knight, however, instead of moving the rook out of danger, Soap takes the bishop, and Termy plays...Bh4??? But that's just a free bishop...right?
icy slides the rook to f9, and Termy plays Bm8+, releasing a discovered attack on the red queen while also delivering a check to icy's king. Soap proceeds to take Termy's green rook and Termy takes Soap's rook:
icy blocks the check with the bishop...after thinking for 16.1 seconds. So now he's down to 0.4 seconds on the clock. Soap is also pretty low on time, with 23 seconds.
While icy is in check, Termy attacks the rook with Nd8, but Soap anticipates this and plays Qn7, followed by Qh7 to recapture. Termy is down to a green queen and knight. If RY can force a queen trade, they will be winning, because while the material is about even, RY share the material while Blue has all the pieces and Green has a single knight. Then, RY can coordinate attacks on Blue to win his material and Green can't do much to prevent it. So Termy tries to hold onto the queen for as long as possible.
icy makes use of the pin on the green king and plays Bxk9, with multiple check threats, so Termy moves the king back to n8. This is now the current position:
Soap takes the knight, and nothing much happens for a couple moves. icy takes a pawn, Soap and Termy trade bishops on h2, and Termy gives icy a check with the queen followed by another check to Soap's king.
After all is said and done, RY are up a bishop and 3 pawns, but the game still continues for another 28 moves, with icy still at 0.4 seconds. But remember when we talked about material imbalances, and how Green losing their queen would result in a losing position? Well, RY secures a queen trade:
Do you see the idea? Solution (played in the game): Qj5 followed by Bk5+, a discovered attack on the green queen, forcing a trade. Termy plays Rk8 in the game, but if it played Rj8 instead, icy can still play Bk5+, and then Soap could capture the rook for free. Since j8 is a dark square, and the yellow bishop is a dark-squared bishop, icy can defend Soap's queen after it captures Termy's rook. icy plays i11 to recapture the knight after the queen trade.
And now, RY are simply winning. All of Termy's material is secluded to the Blue pieces, while Green only has pawns. Combined with the fact that RY are up a knight and three pawns (they lose a couple later, but this does not change the outcome), it's only a matter of time before they can convert—and finally win against the undefeated machines. Termy captures a few more pawns and gives a couple of checks, but RY's rooks are simply too powerful. The rest of the game is some endgame maneuvering that ultimately results in the rooks infiltrating the blue position and checkmating the Terminator (one of the nicest checkmates I've seen lately).
Ladies and gentleman, the once unwinnable robot, genius in its war tactics and ruthless in its attacking advancements, has been defeated.
Of course, it became much stronger with more training and testing and has been undefeated ever since. But we don't talk about that. You see, unlike Arnold, this monster did not peak and then fall off. It just keeps getting better.
Players' Comments
It's been almost two years since the end of that legendary game, yet ask anyone that was spectating at the time and they will tell you how incredibly hype it was to see the decisive conclusion, that proved once and for all, that no matter how incessant an engine's play may be, it is still beatable. So of course I had to get a first-hand account from the man, the myth, the legend himself, @icystun, about his thoughts on the Terminator and its insane exponential improvement over the years:
I don't recall my first game against Terminator but I remember it being possible for the top players to beat it regularly. I thought it was great to play with engines and it is always fun to put on a man vs machine show, especially when the humans stand a chance. In the 4pc old setup, RY has a good advantage to start, so it gives the humans a shot even now.
I remember my latest win. That felt great and it was around that time very difficult to win against the engine.
I've noticed that Terminator plays with no regard for king safety and often misunderstands closed positions (nonconfrontational openings)! A typical engine from the early age of computing. Often having problems with horizon effect when it comes to deep attacks. It makes for enterprising games though.
At some point I had some "bet" to play until I win a game against Terminator. I think we lost 20 in a row before we won that already referenced game...it was mostly a great practice and overall it is a quite close matchup even if the results now are quite lopsided in the machines favor.
Terminator helped to teach me to play better defense and mostly creative ways to defend, often with a sneaky bishop move, often the most counterintuitive way to give space for the king, etc.
— @icystun
What an amazing story!
Here are some thoughts from other prominent players:
@DreamConquerer's thoughts (and POV rankings of engines based on strength). Dream is a strong teams player and an active member of the community.
I'd like to access their databases to look into openings but at the same time I understand that anticheating measures would required spending cash the server doesn't have.
—@DreamConqueror
As for their playing strength, Termy is still the best. I think Titan would be second, then Phoenix. Machina is the laste one for now.
That's pretty much all I have to say about engines at the moment.
@HSCCCalebBrown's thoughts. Caleb is a prominent FFA player and has compiled a lot of data in the past related to player strength, estimated rating, championship winners, and more.
First, not that it completely relates, but I wish we had an FFA engine. It would be fascinating.
So, and this is less of a 4pc specific comment, though it does relate, but I look at top level chess and it appears that engines hurt their
That probably doesn't relate much, because 4pc doesn't have all the infrastructure 2pc has. But that's always my general concern with all this.
More knowledge isn't always better. Don't get me wrong, I'm not saying engines are bad. But what is the place for them, and are they going to help our game or hurt it?
That was one of my concerns with the Old Setup. It's imbalanced; are we going to find something someday, a line that is similar to a forced draw in regular chess, or like a win, or at least major advantage for RY—all with the use of engine analysis?
Right now it's a novelty thing. So, in the future what's the place for it, and is that a good thing?
So I guess what I want to say is that, as we grow and develop all these things, we need to be careful for the future so we don't end up with things like the draw issue...
— @HSCCCalebBrown
@Matttin's thoughts. Matt is currently a 4pc admin and a friendly face known by many in the teams community.
I think engines were a good addition to the game. They changed our ways of thinking significantly. For example, the Terminator evaluation of different starting setups was one of the main argument points that lead to New Standard, but it also created some eccentric opening theory for the Old Setup (hasn't had that many games in New.
What's special is that their strengths are (usually) really predictable compared to players and therefore nice for practice. I hope we get more strong engines but also that they don't make theory too ridiculous.
— @Matttin
And of course, we can't complete the...quoteology(???) without a DISCUSSION(!!!) between twO mORe pRomINeNt plAYeRs (im sorry i couldnt think of any other words so im copy pasting this sentence): a very skillful @ConcreteStreets, and @omatamix, the programmer of PhoenixZero himself. For reference, here is the link (in game chat) to the full conversation they had. Please note that I corrected some grammatical inconsistencies (though the "casual texting" nature of the discussion is kept) and may have reworded some sections to make it more comprehensible (and strictly simplistic) in the given context.
O: overall the move quality of the engine is a lot better: still working to perfect it completely. Phoenix has around 16k lines of code, which is more than I wanted
CS: more time spent than you thought?
O: yes, the first version of Phoenix was only 700 lines of code
CS: now we have to wait for Elsa to come up with her engine
O: yes gonna be a challenge because I struggled xD
CS: how much time did you spend on it so far, what do you think?
O: lost count lol
CS: so what are you going to do with it? perfect it and then never use it like terminator? dude made a great engine but only lets it play like twice a year
O: I want it to learn enough to where terminator losses every match and is impossible to beat. I told chess.com they can have access to the code and if they wanted to implement it in the 4pc variants server they can. I want my engine to run 24/7
CS: what did they say?
O: They said no
CS: who did you message? spacebar (@spacebar)?
O: yes, and Baba (@BabYagun)
CS: that's kinda disappointing actually
O: I would hate to release the code because that would destroy theory
CS: do they just not want to put in the effort to implement or what's the issue
O: people would just play engine lines
CS: you could try weird openings that are considered impossible/bad as of now and find lines that make them work that would be interesting, maybe use it to review puzzles for errors
O: yes Phoenix usually solves puzzles pretty well
CS: but the thing is if it were public to play you would see players playing engine lines anyway, kinda what happened with terminator: played some tricky openings and got copied
— @ConcreteStreets & @omatamix
There are some pretty fascinating takes on 4pc engines that I hadn't actually considered much and can spark some serious but descriptive discussions and debates. Thanks to all of you for providing these interviews!
A Short Discussion on Cheating
With such strong engines like the Terminator, you'd think there'd be a lot of cheating in Teams. But it's safe to say that the closest thing to cheating we have is people like me baiting their 1500 friends into helping them farm +0.5 rating gains every game. Actually no, that's @fourplayerchess, not me. Trust 🤣🤣🤣
This is mostly because these engines are closed-source. They aren't available to the general public for use, and I'm not even sure if you can ask the developers to "lend" you their machine for analysis. As omatamix mentioned in the previous section, he isn't really fond of releasing his engine to the public—the main concern being that, while people originally use the engines to learn and prepare for games, they will sooner or later start playing engine theory only from then on. It's all an example of regression to the mean.
Furthermore, these engine challenges only appear in the lobby once in a while. Phoenix hasn't played in a couple weeks, and neither has Machina. The Terminator hasn't played teams in a very long time (like when Concrete said that Tony only lets the Termy play twice a year), but Tony has additionally decided to test his bot's performance in Duck Chess. Putting it in realist terms...it demolished its opponents pretty badly. But that is a topic for a whole other blog.
Unfortunately, cheating in FFA is easier and therefore more prevalent (through prearranged teaming), and cheating in Atomic is probably the worst that we've seen. The deviant mind is often difficult to understand, but what can we do? Cheating is more prevalent than ever before, with an all-time high of user traffic in 2 player variants. These topics lead to interesting discussion and debate, but they're not all that related to the concept of this blog. Thoughts? Leave your ideas in the comments.
Expectations for the Future
TeamTitan itself is a relatively new engine, and people are still trying to find ways to plan out and program an FFA engine. At this rate, we might be seeing even more engines pop up in a relatively short amount of time. It really only depends on who is willing to devote their time and effort into creating a bot that plays 4pc, and who is willing to play against said bots.
My thoughts on this whole situation? Frankly, I don't see many downsides to new engines. Of course, it would be best to practice a controlled, maybe even restricted environment, to prevent these powerful AIs from crushing the human players, but it's definitely fascinating to see how the human brain thinks differently from an artificial brain that was created by the same person.
Conclusion
In any case, I hope you enjoyed! These are some pretty cool games played by some pretty well known players and strong engines. I think it's impressive that we've managed to produce such strong AI within only a couple years, but could that be a possible weakness in the future? Hopefully we don't get run over by a stream of Terminators. 🤷♂️ And yes, it finally came back to play some more teams games!
And before I leave you, here's a bonus game played by Termy, against @valger2, who is otherwise, with the exception of icy, the greatest teams player of all time. That's all for now! Take care, and stay tuned for another read from Cheese Academy.
