Note: 1 hour, 25 minutes, and 18 seconds of loads were removed from the above encode.
Objectives
- Uses "hardest" difficulty
- Forgoes time-saving glitches
- Forgoes "warps"
- Genre: Board
Story, Methodology, and why is this run so slow?
This is a follow-up to my previous submission for skill level 1 I made 2 years ago, where I said in the discussion thread that "I would've picked the harder difficulty if the computer wasn't so damn slow!" In that thread, I mention a test game where I played on Skill Level 6 and made every move recommended by Stockfish 12 at 35 depth to see how long a "perfect" game would be. During the game, the Intellivision AI thought for a total 7 hours, 12 minutes, and 54 grueling seconds. Since Bizhawk tops out at ~600% speed unthrottled on my laptop, this game still took about 1 hour, 10 minutes to play in full, so putting in the effort to craft a faster game through trial and error would be an incredibly mind-numbing experience...
So that's exactly what I did... I painstakingly improved the above game by 5 hours, 45 minutes, and 58 seconds through sheer trial and error... the only practical way to improve this without losing one's sanity would probably be through a similar method to negative seven's A2600 Othello TAS, where games could be modeled through an automated process. In this final result, I didn't actually use a chess engine to recommend moves as I eventually figured out and exploited the kind of positions the Intellivision either thinks briefly for or has weaknesses in evaluating, which I will explain in depth later.
For those of you who are not impressed and think an hour and a half chess game must be suboptimal and easily improved, I encourage you to play this casually. I guarantee you'll run into a situation where the computer thinks longer than the entire length of this TAS for a single move. Believe me, such positions are common. At worst, the computer can take days to think of a move. That's right, DAYS. And it's really not too surprising once you learn the game only had 2147 bytes of RAM to work with.
Difficulty/Goal Choices
There is an absolutely overpowered time-saving glitch I did not use where you can essentially force the computer to make whatever move you want it to, but explaining that in full is beyond the scope of this submission. See the TAS I submitted alongside this one which takes full advantage of the glitch.
Skill Level 6 is actually the second-hardest difficulty, so what gives? Well on Skill Level 7, the computer will "think" indefinitely, not moving until the "Force Move" button (lower-right side button) is pressed, which forces the computer to make whatever move it has so far deemed the best. Not pressing the "Force Move" button is what I mean by the "Forgoes Warps" objective as it's kinda like warping through the AI's decision tree. Not to mention, the instructions manual states on page 12, "When you press Force Move, you are not playing at the skill level you chose", a sentiment which I agree with. With this restriction therefore, Skill Level 6 can essentially be considered the hardest difficulty.
Routing
Very important!The goal is not necessarily to checkmate in the least number of moves but to deliver the checkmate that makes the computer "think" the least. Shorter "thinking" times are achieved by keeping the position simple. Giving the computer few reasonable options and not attacking with a lot of pieces simultaneously is a good way to minimize thinking times. But the BEST way to reduce think times appears to be trading or just straight up blundering pieces. If the computer sees a piece where capturing will grant a material advantage, it'll take without much thought. This is evidenced in the game summary where the shortest think times besides those in the opening and before imminent checkmate are ones where Black captures pieces. Evidently, the Intellivision is a very materially-focused chess engine with not as much focus on tempo or positional outcomes, and are flaws which can be exploited.
We also saved some time by hiding the clock displays, which is done by pressing the "8" key twice on the right controller. This frees up some additional RAM to be used for computer thinking and reduced the total time by 54 seconds.
Situations to avoid when trying to play fast
I've noticed the AI tends to think for over 10 minutes, or even over 24 hours sometimes, during the following:
- Considering a developing move after move 3-4
- Complex mid-game
- Highly positional endgame
RNG
Moves are NOT determined stochastically. The computer's response to a position at a given skill level will be the same every time.
Game Summary, Commentary, and Alternate Moves Discussion
| Move | White | Black | Black's Time | Move Time |
|---|---|---|---|---|
| 1 | g3 | Nf6 | 3:03 | 3:03 |
| 2 | Bg2 | e5 | 7:29 | 4:26 |
| 3 | c4 | Bc5 | 15:06 | 7:37 |
| 4 | d3 | O-O | 29:22 | 14:16 |
| 5 | Bg5 | Bxf2+ | 33:13 | 3:51 |
| 6 | Kf1 | h6 | 41:31 | 8:18 |
| 7 | Bxh6 | Bxg1 | 46:45 | 5:14 |
| 8 | Bg5 | Bd4 | 55:10 | 8:25 |
| 9 | Ke1 | Bxb2 | 58:12 | 3:02 |
| 10 | Rf1 | Bxa1 | 1:02:27 | 4:15 |
| 11 | Rxf6 | gxf6 | 1:05:08 | 2:41 |
| 12 | Be4 | fxg5 | 1:07:02 | 1:54 |
| 13 | e3 | Nc6 | 1:17:56 | 10:54 |
| 14 | Qh5 | Re8 | 1:24:47 | 6:51 |
| 15 | Qh7+ | Kf8 | 1:24:52 | 0:05 |
| 16 | Qh6+ | Kg8 | 1:25:45 | 0:53 |
| 17 | Bh7+ | Kh8 | 1:25:49 | 0:04 |
| 18 | Bg6+ | Kg8 | 1:25:55 | 0:06 |
| 19 | Qh7+ | Kf8 | 1:26:01 | 0:06 |
| 20 | Qxf7# | 1:26:02 | 0:01 |
1-4. The primary goal of the first four moves is to get black to castle king's side as fast as possible as black has a certain... weakness... in preventing checkmates over there. We'll get to that later. Our sequence of g3, Bg2, c4, d3 prompts black to castle on move 4, which is the earliest possible. You'll notice that after move 4, both of white's bishops are now able to immediately jump into the action which is a great advantage.
- g3, Bg2 were found to be the only first two moves which persuade black to work directly towards castling while also getting a piece, our light-square bishop, ready for action.
- c4 seems like a waste of a developing move, and indeed, it seems like it provides no benefit for the rest of the game. However, this is the key move that coaxes black into playing Bc5, which leads to five additional moves from this bishop by black with relatively little thinking time. Assisting to that end, c4 also weakens the pawn on b2 and the rook on a1, because now black sees that there's no defensive pawn to c3 move if black posts its bishop up on d4. It might seem odd why we would want to invite such activity from black's bishop but that will be explained when we get there...
- Let's say we played c4 on move 1 or 2 instead of 3. It would seem most sensible to play it on move 1 as that would appear to save on cursor travel time. However, Black would play d5 instead of Bc5, which both attacks our c pawn and contributes nothing towards getting black to castle. We can do better than this.
- e5 is also not a viable developing move as this once again prompts d5 from black.
- d3 is pivotal to getting our dark-square bishop in the game but can only be played after black plays Bc5, as doing so beforehand will allow Bb4+. On this square, black will not have the same activity with this bishop for the rest of the game, which we will steer away from.
- Now let's look at some knight moves.
- Playing Nc3 on any move between 1 and 4 will cause black to respond with d6, protecting the e5 pawn. While the think time for d6 now is relatively short, not only does this not contribute towards black castling, but later on it will discourage black from moving Bd4, which as I explained earlier, we would like to happen.
- Playing Nf3 on moves 1-2 will prompt d5 instead of e5 and as I noted earlier, is an inefficient move for us.
- Playing Nf3 on moves 3-4 instead causes black to go on the attack with e4! Yeah, let's avoid that one too.
5. Once black castles, we immediately start putting pressure on the king's side by moving the light-square bishop on g5 to attack the knight (which Stockfish says is a blunder but shhhh... don't worry about that!). This knight is a key defender for black and prevents a myriad of mate threats, so we must get rid of it. The fact that it is pinned to the queen helps it stay put. It also coaxes black to play h6... but not yet. Black ignores this for now and goes for an admittedly nice tactic with Bxf2+. It might seem like we're losing tempo by allowing this check, but did you notice something? By taking our f pawn, black just opened up the f file, which we will use to our advantage later.
- Playing Nc3/Nd2 instead of Bg5 yields a d6 response from black after 10/12 minutes of thought which isn't a great start and once again, discourages Bd4 later on.
- In addition, playing Nd2 gets in the way of developing our light-square bishop, so Ne4 must come before Bg5. After 11 minutes of thought, Nxe4 by black, then after Bxe4 by white, black thinks for 19 minutes before playing c6, and suddenly, too much time has been wasted, making this a suboptimal pattern even though black's knight on f6 has already been removed.
- Playing Nf3 instead of Bg5 prompts black to play Nc6 after 14 minutes of think time. We can also do better than this.
- From this move up until move 9, I also experimented with pushing the h pawn forward to see if I could get any favorable outcomes weakening black's defenses. In all attempts, it is too slow with three moves required just to get the pawn on h6, moves which are better spent elsewhere.
6. The king must be moved to f1 because the other two options are just bad as explained below. Black now responds with h6 to kick the bishop out.
- The idea of Bxf2+ appears to be that if white plays Kxf2, then black can play Ng4+, followed by Qxg5, gobbling up our bishop with a positional advantage and be up a pawn. So we can't take that bishop. In fact, capturing black's dark-square bishop on any move would slow the game down considerably. Look back at the game summary above and notice how black will move their dark-square bishop for the next four moves in a row, and also notice how briefly the computer thinks for each of those moves. Black is very sure taking the knight on g1 and then the rook on a1 is a winning continuation. If black's bishop was gone, it would have to go into a deep think for some other move. Giving free reign to black's dark square bishop is perhaps the most critical time save in this TAS. So, while black is distracted doing their little tactic and taking pieces of no value to us, we continue working towards a checkmate.
- Kd2 is even worse as after Ne4+, white can do either Kc2 or dxe4, neither of which are stopping Qxg5.
7. If black were to capture on g5, that would open up the h file and weaken the king's position. But we do even better by capturing the h pawn ourselves, and now if black recaptures, the g file is the open one and is even weaker for black! Surprisingly, black does NOT recapture and instead takes our knight on g1!
- Capturing the knight on f6 with our bishop between now and move 10 is bad news, as this will prompt Qxf6, which not only brings the queen towards defense of the black king but also threatens check or discovered check. We need a second piece attacking that knight on f6 to prompt something like gxf6 instead...
- It's a shame we lose our g1 knight as with just two moves required to post up on the g5 square, it could play a pivotal role in delivering a checkmate. Problem is, not only does our bishop want to attack the knight on f6 from the g5 square, but moving our knight does nothing to address black's problematic knight on f6, which is the more pressing issue. Our g1 knight therefore better serves as a punching bag for black to just take with minimal thinking time, and it doesn't even cost us a move!
8. We save our dark-square bishop from recapture by moving back to g5, once again pinning the knight on f6 to the queen. Black meanwhile quickly sees that our rook on a1 is vulnerable and decides to attack it... oh no... not my rook who will contribute absolutely nothing to this game!
- We also have the option of capturing the pawn on g7 which would open up both g and h files to attack the black king. This prompts the black king to recapture but now... nobody's attacking the knight on f6. Spoiler but we're going to be opening up both files in a more efficient way later on.
White's moves 9-14 can be done in many different orders. I tested all possible permutations. All except two prompt black to play d6 instead of Nc6, which requires more thinking time and all those are therefore suboptimal. The other variation which gets Nc6 switches the move order on move 12 and 13 to e3 followed by Be4, but is still 21 seconds slower for some reason.
Let's now analyze moves 9-11 specifically:
- So, the big tactic to oust the knight on f6 is to move the king back to e1, Rf1, then Rxf6, made possible by black opening up our f file earlier. But is there a better way? I mean, that's two king moves in this game. It's hard to believe this TAS is fast with two moves like that but remember, it's all about keeping AI thinking times short, and the king moves stay true to that.
- The above sequence requires three moves to execute. Well, our knight on b1 can get to f6 in three moves in three different ways so let's take a look at each:
- Moving the knight to c3 causes a big think because now the AI has a choice to make! Oh no! Do I take the knight or the rook? Black ends up taking the rook but the think time is so long its not even worth analyzing the Nd5 or Ne4 continuations.
- OK so what about the Nd2, Ne4 variation? Well, unlike when the rook attacks the knight on f6, the knight on f6 now must evaluate whether it's worth taking the knight on e4. It's not worth so it must think... a while... for a counterattack and eventually settles on d5, but too much time has passed for this variation to be optimal.
- The above sequence requires three moves to execute. Well, our knight on b1 can get to f6 in three moves in three different ways so let's take a look at each:
- I also tested what would happen if Bxf6 instead of Rxf6 and depending on your move order afterwards, you'll get black to play either d6 or Bd4 which are both slower than Nc6.
Let's now analyze moves 12-14 specifically:
- These are mostly positional moves that get the bishop then the queen in an active position to prepare for checkmate. You might notice on move 12 that we just let the pawn on f6 capture our dark-square bishop for no reason? Yeah! That thing is no longer of any use to us so no point in saving it. We have more important things to do with that tempo! While this closes the g file back up, it is of no use. The g pawn has advanced too far and there is no stopping the queen from getting all up in the king's business.
So what did black do on moves 9-14? Well, after wasting the first two moves picking up a pawn and a rook, black follows this up by capturing on f6 then g5, and so far this is looking really good for black. I mean, they just picked up a net 11-point advantage in the last 4 moves! But look at black's bishop on a1... look at the 8 pieces in the upper-left that black hasn't even moved yet! They are all way out of play and the king is about to get mated. Oh no but black plays Re8 on move 14 and now the king has an escape path! Damn, we came so far and yet I can't believe black is about to get out of this scot-free. Unless...
15-20. Because of black's last rook move, the King is now able to escape checkmate... unless we trick it by somehow getting the king to move towards the corner. When I move the Queen to h6 on move 16, for some ungodly reason the king moves to g8 instead of e7, instantly losing the game! I guess black didn't want to blunder the g5 pawn by moving to e7, but dude... would you rather blunder a pawn or your king? Lol. The trickery and checkmate process takes a bit of piece shuffling—a total of six moves—but black has limited move options and therefore quick think times, so it is absolutely worth it. And finally, on move 20, the black king has fallen. And so too has my sanity.
Chess.com game review White was -18.7 at one point. What a comeback!!
Piece Movements
In addition to all the focus on think time minimization, there are also some speed strats going on with how we move pieces. A piece or square can be selected even if the cursor is on an infinitesimal portion of the square's right or bottom edge, or the bottom-right corner. This allows us to set all our pieces down quite early, and gives an advantage to picking up pieces if they're moving down or to the right.
The cursor becomes active for the player's next turn from the moment the "gong" sound plays to denote the computer has decided on a move, even though the cursor cannot be seen yet. This is why the cursor is always in the perfect spot when it becomes visible.
The instructions manual says this about the lower-left side button:
Before move is completed, move piece back to last position
For some reason, you can press this button the moment the computer selects a piece to instantly snap it to its destination square, instead of watching the computer mosey its piece on over to the square for a couple seconds time save per move.
Playing as Black?
The (8) key on the left controller allows the player to start as Black instead of White. I didn't bother testing any games as black as by going second, as you're immediately giving the computer an extra turn to think for which loses a lot of time.
Suggested Screenshot Any frame after the "Checkmate" message appears.
Suggested Publication Notes
The year is 1982. Chess engines are in their infancy. There is no Stockfish, AlphaZero, or Deep Blue. If you want to play chess against a computer in the comfort of your own home, one of your options is USCF Chess for the Intellivision. With only 2147 bytes of RAM, prepare to wait hours or even DAYS for the computer to make a move on the highest skill level. And even then, the move the computer decides on could be a bit questionable.
Minimizing computer thinking times is of utmost importance in this type of TAS. By keeping the position simple and tricking the computer into making some dubious moves, Winslinator beats the hardest skill level in just under an hour and a half.
nymx: Claiming for judging. I'm really going to take a deep look at this one. Hang tight, it will be a moment.
nymx This particular chess submission, was the one that I really wanted to dig into and find improvements....but I couldn't. These older 8-bit machines were venturing into the beginnings of chess AI, where code was small, memory was limited, and they didn't have complicated calculations as modern games do. So...the fastest way to beat this game, is to know the CPU's weakness and exploit it. This TAS does exactly that. I have personally experienced the weakness of early computer programs and consoles, thus recognizing it with this movie. With as slow as this game ran, I applaud you for sticking with it.
Playing "Level 6" satisfies the goal of "Full Completion", where no major skip glitches are used. Excellent job.
Thanks to ikuyo for your review.
Accepting.
despoa: Processing...