"The decisions you make in the game should be agonizing," he replied.
September 23, 2014 12:28 PM   Subscribe

What is to stop you using a bot?
posted by bhnyc at 12:35 PM on September 23, 2014

The website makes it look like a John Hodgman text adventure.
posted by Strange Interlude at 12:39 PM on September 23, 2014 [4 favorites]

A rake (fraction of the amount bet that house keeps) of 10% is super high. I bet he'll have to decrease that to attract regular players.
posted by dfan at 12:42 PM on September 23, 2014 [2 favorites]

What is to stop you using a bot?

Would a bot help? Is there clearly an optimal strategy? I think probably, but it's hard to tell for sure...
posted by mr_roboto at 12:43 PM on September 23, 2014

wow I am really fascinated by this game - i look forward to the forks that people release that don't require cash.

WRT the bot, this seems a lot like the previous comment by painquale about the Rock Paper Scissors bot competition, which required a super meta strategy, but not one that a human could not carry out themselves. I think mastering the meta will be just as much a part of the game as solving the puzzle itself.
posted by rebent at 12:47 PM on September 23, 2014 [1 favorite]

The meta will be tough if you're playing a new random person every game.
posted by dfan at 12:50 PM on September 23, 2014

WRT the bot, this seems a lot like the previous comment by painquale about the Rock Paper Scissors bot competition, which required a super meta strategy, but not one that a human could not carry out themselves.

Yeah, but RPS gives you super-minimal information (just your opponent's previous throws, which may or may not mean anything) and you have very limited choices (one decision, from three options, at each throw). It seems that in CM, with the field of numbers and the ability to select two rows in addition to a wager amount, there's a lot that can be done analytically before you even start thinking about meta-strategy.

Just being able to calculate the probability of each possible final score and make a wager decision based on that would be really helpful, it seems. But maybe not...
posted by mr_roboto at 12:56 PM on September 23, 2014

Would a bot help? Is there clearly an optimal strategy? I think probably, but it's hard to tell for sure...

I don't think there is. I think the game is basically iocane powder, without any chance for the immunity trick.
posted by aubilenon at 12:56 PM on September 23, 2014 [3 favorites]

Yeah, okay, that's definitely going to replace poker.

posted by Naberius at 12:57 PM on September 23, 2014

I take it back. There are sometimes clearly dumb plays and there are sometimes clearly right plays. That means there's something to it. But I think a lot of it basically is just luck about who guesses what the other one will do better.
posted by aubilenon at 12:59 PM on September 23, 2014

This is just two normal form matrix games played at the same time, with flipped roles. Solutions to this problem have been known for a while. (For example, see page 55 of this pdf for a linear program formulation). Note that, like in rock paper scissors which can be expressed as such a game, the optimal strategies are in general random (like in rock paper scissors, where the optimal strategy [in a Nash equilibrium sense] is to randomly pick from RPS with equal probability).

This game will be overrun with bots because:
1) This is a well-studied game theory problem with an easily computed Nash equilibrium
2) Knowing that your opponent is playing the Nash equilibrium strategy doesn't let you beat them, it only lets you tie
3) Since the NE strategies will almost always be mixed strategies (i.e., randomized), it will be very hard to prove that anyone is using a bot just based on their behavior
posted by Pyry at 1:00 PM on September 23, 2014 [19 favorites]

From the Kotaku article: "It was all legal, he assured me"

Pretty sure the UIGEA covers this too.
posted by thewalledcity at 1:01 PM on September 23, 2014

i wonder if there is any kind of Elo match making system.

What if you had 2 instances of the game running, with bots, seeded with a bunch of money and playing with minimum bets, using the moves of opponent 1 vs opponent 2 and vice versa.

Eventually, one might start winning more than the other, which presumes that the loser bot will start getting matched against weaker opponents and the winner bot stronger. When there is a large enough deviation, the bets are skewed to favour the loser bot (since it will be using the moves of a stronger opponent). When a big deviation occurs, a large bet occurs.

Since you don't have to worry about random elements, your bets can be made based purely on the strength of the opponents.

Of course, if there isn't a reliable deviation of more than 10% (the rake) this might now work...

Also, since I'm not really that smart, there are probably several other reasons why this wouldn't work.
posted by Reyturner at 1:03 PM on September 23, 2014 [1 favorite]

The game board is randomized, though the board always is a "magic square". For this reason at least, you can't really use the moves of one player against another via a pair of bots.

Unlike, for example, chess.
posted by rustcrumb at 1:08 PM on September 23, 2014 [1 favorite]

posted by Reyturner at 1:09 PM on September 23, 2014 [1 favorite]

Not sure a game can replace online poker if the title sounds like that of a minor Neo-Reactionary Wordpress blog.
posted by Rustic Etruscan at 1:11 PM on September 23, 2014 [1 favorite]

This doesn't look like any fun at all
posted by Ray Walston, Luck Dragon at 1:13 PM on September 23, 2014

What's the sinister anagram hinted at in the review? MORTICIAN DUEL? ORCA UNLIMITED? ACUTER MIND OIL? I RECLAIM DONUT?
posted by Iridic at 1:34 PM on September 23, 2014 [3 favorites]

posted by rustcrumb at 1:49 PM on September 23, 2014 [5 favorites]

Would you like to SUMMON ASTAROTH to your bedchamber using nothing more than INEXPENSIVE ITEMS found in your kitchen pantry?

Lemme know when he's up to summoning Azathoth with an opera.
posted by JHarris at 1:50 PM on September 23, 2014 [1 favorite]

This game will be overrun with bots because:

[excellent reasons follow]

I'm actually quite surprised that Rohr hasn't realized this himself. Even without any real understanding of game theory, I know enough about software engineering to realize that this decision space could be fully mapped out very quickly. If that isn't enough to give you an optimal (or at least, a really good) strategy, then you're not playing a game of skill.
posted by Edgewise at 2:01 PM on September 23, 2014 [1 favorite]

the optimal strategies are in general random

Well, this guarantees you an average 50% win rate, no matter what your opponent does. Since the house gets 10% that isn't very good.

But if you can recognize someone playing a non-random strategy you can exploit it. But then you open yourself up to being fooled. Maybe the equilibrium state is everyone just gives up on the game. Except dumb people would still play, so it might make sense for smart people to try to take their money, right?

Anyway the real depth to this game, if there is any, is in the betting and bluffing. Like getting the best hand in poker isn't how you win at poker.
posted by aubilenon at 2:14 PM on September 23, 2014 [1 favorite]

I could imagine him just publishing the Nash equilibrium and saying it gets you even odds but negative expectation. Sort of like how casinos sometimes give you basic strategy for blackjack.
posted by vogon_poet at 2:21 PM on September 23, 2014

My last comment sounded more supportive of this game than I really am.

My actual opinion is: If you play randomly, the game is equivalent to a slot machine. If you play strategically, then the matching system is equivalent to a slot machine.

So basically it should be legislated the same.
posted by aubilenon at 2:48 PM on September 23, 2014

An additional note:
The game is played in three rounds, where in each round you can't pick the options you choose in the previous rounds. So that complicates things a bit.

But I suspect that it disproportionately complicates the game for humans, and not for computers. Now your bot has to do some kind of minimax tree where each node's expected value comes from the Nash equilibrium for the choices available at that point, but given the limited depth (only three rounds) and limited choices (only 30 choices in the first round [6 columns choose 2, order matters]), this is a rather small game tree for a computer to deal with. The imperfect information (you only know one of the other's player's choices) grows the tree a bit, but not past the realm of feasibility.

I don't fully understand how the betting is supposed to work, but it could potentially add another wrinkle. So maybe all the interesting gameplay will come down to betting.

I could imagine him just publishing the Nash equilibrium and saying it gets you even odds but negative expectation.

So in RPS the Nash equilibrium is not very interesting, because not only does it have zero expectation, but also it can't win against any strategy (it always ties, in expectation, against every strategy).

But this isn't true of all games/matrices: for example, consider the game "rock, paper, scissors, dull-scissors", where dull-scissors works like scissors, except that it loses to regular scissors. Obviously, there is no reason to ever pick dull-scissors over regular scissors, and the Nash equilibrium reflects that: it's still to pick randomly from RPS with equal probability (never picking dull-scissors). This strategy still has a value of 0: the other player can counter by also doing the same thing and you'll tie. However, the NE has a positive expectation against strategies that are stupid enough to ever pick dull-scissors. So, for example, the NE would have a positive expectation against the strategy to pick randomly from all four options.

I don't have any intuition as to what the expectation of the NE strategy vs. say, random choice is on a random 6x6 magic square matrix, but it would be easy enough to estimate by just generating such matrices. That would give you some idea of how profitable you might expect such a bot to be.
posted by Pyry at 3:08 PM on September 23, 2014 [2 favorites]

It would be kind of surprising if he hadn't considered the bot angle. Look at the academic work on his homepage. He's thought about stuff like using neural networks to play go. He's certainly alert to the idea of a solvable game; I'm sure he knows about poker bots.

Now as to whether any of his thinking on these issues is any good, who knows? But he must have thought of it....
posted by mr_roboto at 7:07 PM on September 23, 2014

I too was puzzled at the easily computed mixed strategy nash equilibria here. But apart from how that impacts the play, I wonder how it impacts the legality. That is, even if the game itself is not one of chance, if the optimal strategy is mixed -- ie, randomized -- won't every game actually be, effectively, a game of chance? Maybe this is all just a performance piece to show how MNE mean that all games are games of chance, and therefore there can be no principled distinction between regulating games of chance and any other game.
posted by chortly at 7:19 PM on September 23, 2014 [1 favorite]

I wonder if you could implement this as in Lua as an add-on for WoW? They shut down a couple of "casinos" because they were being run unfairly. With no chance element involved in this, it would let people gamble gold while waiting around in the game.
posted by ob1quixote at 9:07 PM on September 23, 2014

So what's the deal? Did he just not know about game theory when creating this game? Or, if he does know it, what's his angle?
posted by rebent at 5:42 AM on September 24, 2014

The deal is that I'm reasonably well-versed in both applied game theory and AI programming, and I've already tested it with some simple bots that I coded up myself.

Yes, this is essentially a game played on a payoff matrix. But because multiple picks are involved on the same matrix, a simple mixed strategy isn't sufficient. In my testing, greedy beats a uniform random strategy, and straight-up minmax (in a decision tree with 518,400 possible leaves) beats both greedy and uniform-random. Since it's a magic square, each column has the same average payoff of 18.5, but the multiple rounds played on the remaining columns make the decision tree relevant when calculating a mixed strategy.

Of course, added into this are the between-pick, no-limit betting rounds. Writing a bot that bets, raises, and folds well is a rich problem.

But yes, poker bots exist, chess bots exist, and RPS bots exist.

I expect that bots will exist for this game.
posted by jasonrohrer at 10:49 AM on September 24, 2014 [11 favorites]

It seems like if you want people to keep playing when they're losing, they need to feel like it's a game of luck, because otherwise it feels like they're just paying someone to tell them that they suck.

So two ways I could see this failing are: 1) They don't convince enough people that it's a game of skill, and it becomes illegal. 2) They convince too many people that it's a game of skill, and then nobody plays.

My money's on the first one.
posted by aubilenon at 11:34 AM on September 24, 2014

Well, I certainly did not expect the subject of my first-ever FPP to show up and comment. PRETTY NICE WEDNESDAY FOR ME.
posted by joelhunt at 12:57 PM on September 24, 2014 [3 favorites]

I have a question, preferably for Jason if he's still reading this thread... My question is, if this is legal, how come we don't hear about people staking money on CoD and SC2? It seems like it would be easy to build an auxiliary website where you can bet on winning, which could verify results through whatever scoreboard API is available. For that matter, there's nothing stopping publishers from setting this up on their own. So why hasn't it happened yet?
posted by Edgewise at 1:13 PM on September 24, 2014

Well, it has happened already, sorta.

First case, fantasy sports, which are currently taking in $15 billion a year from 32 million players in the US alone.... that's like 1 out of every 10 US citizens. My guess is that it covers about 50% of the 30-50 male demographic. There are entire TV programs devoted to fantasy football now, it's that big.

Second case, "Skill gaming," which had a small bubble a few years ago. Some residue from that wave still exists, like:


This is actually run by King, the company that makes Candy Crush. In fact, Royal lets you play Candy Crush against another person for real money. King originally started out as a skill gaming company back in 2003. Read about the history here:


But their idea never seemed that interesting to me.... like, take an existing 1p game (e.g., Bejeweled) and rig stakes for two players around it (by giving them the same random seed and seeing who gets the higher score on that board).

What has always interested me is designing a new game, from the ground up, around this model. Poker wasn't designed, but what if we had to start from scratch and design a new game in that family?

But anyway, if you deposit $10 on Royal and play around with it, you'll find a pretty compelling experience there (Gospo's Adventure is a very nice, deep matching game).

As for why Valve or Blizzard hasn't tapped into this space, well... it's messy. I've got to collect W9s and send out 1099s to any player that profits more than $600 in a given year. I've got to pay attention to a changing legal landscape in 50 states.

If they're raking in more money than they know what to do with by selling you virtual hats or virtual card booster packs, and they never have to send money the other way, why would they bother?

But I'm a game designer, not a profit maximizer. I don't want to sell you a hat. I want to design a game that does something new and interesting to you.

As for why some third party hasn't screen-scraped Counter-Strike high score tables to add cash stakes to games.... well, maybe no one has thought of it! Go for it.
posted by jasonrohrer at 9:06 AM on September 25, 2014 [6 favorites]

Thanks, Jason! This gives me a much better idea of what you're really trying to do. Your rationale for why Blizzard etc. hasn't done something like this makes sense. I'm still not sure why some third-party hasn't tried to do it (maybe they would get a C&D from Bliz or whomever), but like you, I wouldn't find it very interesting to implement something like that myself.

By the way, the game itself looks very interesting. I very much admire the fact that you are able to come up with an interesting game that is purely numeric. I'm sure you're aware that something like this can be played in a non-digital format as well; if it catches on, you might consider publishing a book of randomly-generated grids like Sudoku puzzles, along with special tokens that allow players to select their columns simultaneously.
posted by Edgewise at 3:18 PM on September 25, 2014

Yeah, the offline version would work, along with the selective reveal of hidden info. The columns we each pick for each other would be shared knowledge, exposed after we picked them (turning cards face up after we both pick), while the columns we pick for ourselves would be committed via face-down cards played out, but only revealed to our opponent at the end.

I think the presentation would suffer somewhat, because players would have to mark columns behind privacy screens and keep track of what numbers are left in their heads, as well as computing possible opponent scores in their heads. The computer version does the arithmetic and combinatorics for you, because I feel like doing that in your head in a comprehensive way is out of reach for most people and not particularly interesting even for the people who can do it (so people would code up "wing man" software to help them... might as well build the wing man into the game).
posted by jasonrohrer at 10:58 AM on September 26, 2014

Holy crap, it's Jason Rohrer! I love the shit you make, man.
posted by cortex at 8:03 AM on September 27, 2014

« Older Have a Guinness (Latte) when you're tired   |   Age eclipsing gender as Canada's major income gap Newer »

This thread has been archived and is closed to new comments