Follow Slashdot blog updates by subscribing to our blog RSS feed

 



Forgot your password?
typodupeerror
×
AI Software Games Technology

An AI Learned Magic: the Gathering, Now Creates Thousands of New Cards 104

merbs writes: Reed Milewicz, a computer science researcher, wowed a major online Magic: The Gathering forum when he posted the results of an experiment to "teach" a weak AI to auto-generate Magic cards. Milewicz had trained a deep, recurrent neural network—a kind of statistical machine learning model designed to emulate the neural networks of animal brains—to "learn" the text of every Magic card currently in existence. Then he had it generate thousands of its own.

He shared a number of the bizarre "cards" his program had come up with, replete with their properly fantastical names ("Shring the Artist," "Mided Hied Parira's Scepter") and freshly invented abilities ("fuseback"). Players devoured—and cheered—the results.
This discussion has been archived. No new comments can be posted.

An AI Learned Magic: the Gathering, Now Creates Thousands of New Cards

Comments Filter:
  • This seemed cooler than it actually is, as practically everything the program generates is completely nonsensical. As such, the end result does not seem special compared to everything else "AI's" have supposedly created in the past.
    Hopefully the randomness hits home a couple of times and gives someone actually useful ideas.

    • by nine-times ( 778537 ) <nine.times@gmail.com> on Friday June 12, 2015 @12:38PM (#49899113) Homepage

      I'm disappointed for entirely different reasons. I read, "An AI Learned Magic..." and thought, "Wow! What could that mean? Did it learn how illusionists perform their tricks? Are they claiming it somehow learned real magic? This should be interesting!"

      And then I continued reading.

    • by div_2n ( 525075 )

      Sounds human-like to me.

    • by mythosaz ( 572040 ) on Friday June 12, 2015 @12:59PM (#49899215)

      No. It made nonsensical cards early into its learning process.

      Later on it made cards like this:

      Light of the Bild
      2WW
      Creature - Spirit
      Flying
      Whenever Light of the Bild blocks, you may put a 1/1 green Angel creature token with flying onto the battlefield.
      2/2

      ...which are pretty good.

      • by xevioso ( 598654 )

        Well it's a good start.

        But...

        There are no Angels in green. They are all white, black and maybe red.

        Cards like this only work as part of the whole if the rest of the set takes cards like this into account. Otherwise it's all random and unbalanced.

        • Well it's a good start.

          But...

          There are no Angels in green. They are all white, black and maybe red.

          Are you sure about that? [wizards.com]

          • Well, he phrased it poorly. There are multicolored Angels that include Green (they're all also white, which has always been the preferred Angel color). There are no mono-colored Green Angels; there are three in black (and last seen with the reprint of Fallen Angel in Eighth Edition), and one in red (Akroma, Angel of Wrath) and one in blue (Illusory Angel). White, meanwhile, has *84* mono-colored angels.

    • This seemed cooler than it actually is, as practically everything the program generates is completely nonsensical. As such, the end result does not seem special compared to everything else "AI's" have supposedly created in the past. Hopefully the randomness hits home a couple of times and gives someone actually useful ideas.

      I watched a demo (can't find it now) of TempleOS [slashdot.org] where... God spoke to you!

    • This must be the first time the phrase "seemed cool" has been used in reference to Magic the Gathering. :)

    • Yeah, this sounds like little more than a Markov bot. You can find them online. No "AI" needed.
      • by Megane ( 129182 )

        Just think of the possibilities of Bayesian MTG cards... now instead of being full of GET TAXI CAB IN INDIA, the /. submission queue could be turned into an entertaining and enjoyable* game!

        * for certain values of entertaining and enjoyable that may not apply to people who don't like CCGs

    • I was actually pretty impressed at how close it came to creating sensible cards, and pretty funny when it made small errors that made the cards absurd (such as the card with, "At the beginning of each player's upkeep, sacrifice a white Zombie creature").

      Getting a computer to generate understandable language is an extremely difficult problem, and all neural networks have an issue with long-tail errors (that is, a small fraction of the results are always ridiculously inaccurate, no matter how good your neural

    • This seemed cooler than it actually is, as practically everything the program generates is completely nonsensical.

      Partly because of this, I doubt that there is really much "AI" going on here. To me, this looks very much like simple probabilistic Markov chaining, with maybe a couple of rules to demark meaningful game elements that can be treated as a unit.

      A bit more than that, maybe, but not much.

    • I agree. There are VERY basic rules and patterns that would be one line of code that would have improved it immensely. This was practically random with almost no hope of purposely coming up with a valid sounding card. Something like black cards can't do direct damage to the other player without a secondary effect or trigger condition. That's a universal law of MTG and yet it prints a black lightning bolt.
  • Sigh. (Score:4, Funny)

    by pushing-robot ( 1037830 ) on Friday June 12, 2015 @12:17PM (#49899003)

    This is exactly why we need a moral framework for AI development.

    • Indeed. Useless MtG cards are an existential threat to humanity...
      • that's not stoping Wizards from printing crap like Scion of Ugin and Comet Storm...

        *grumbles*

        Somewhere out there, a tree is working tirelessly to produce air for us to breathe, then it gets cut down and turned into crap bulk cards.

        • by xevioso ( 598654 )

          Uh, both of those cards are fine for casual. The reason Magic is awesome is cards the Spikes out there consider "jank" can be a lot of fun for casual magic players.

          • I think casual is where it gets less fun because if one casual player in a pool of several casual players pulls better cards than everyone else, especially at common and uncommon, pulling a Comet Storm in your rare slot in a pack of Modern Masters 2015 is that much worse.

            I don't think Wizards is just going to sit by and let this happen though. My huge hope is that with standard going to two, two set blocks coming soon with Origins, this is going to make WotC R&D tighten up sets and have less chaff.

            I thi

    • by Anonymous Coward

      Oh shut up.

    • by tomhath ( 637240 )
      Yea. We'll get on that as soon as we attain World peace, end hunger, and eliminate all disease.
  • So with all the recent fuss over AI and some respectable folks being scared to death of it, I happened to stumble on this great article on waitbutwhy:

    http://waitbutwhy.com/2015/01/... [waitbutwhy.com]

    It's a long two parter, but well worth the read. If you want the tl;dr part, skip to part 2 and search for "Robotica". With that in mind, we're going to end up with a planet of mile-high stacks of Magic: The Gathering cards.

  • Disappointing (Score:5, Informative)

    by vadim_t ( 324782 ) on Friday June 12, 2015 @12:51PM (#49899185) Homepage

    What I thought from the description is that a neural network was taught how to play Magic and somehow generated new cards by trying to play with them. Think for instance of a program that tries to come up with a new chess piece by coming up with a movement pattern, playing games with that piece and trying to figure out whether it'd be useful or interesting to play with.

    This on the other hand looks like something like a markov chain generator. Amusing nonsense that can give humans fun ideas.

    • When I started reading, I thought it might be about the way the manufacturers keep releasing new cards to rebalance the game. An episode of Extra Credits on YouTube talked about how they constantly fiddled with the game so that there was always a new potential super-tactic to learn, but after a while it would no longer be quite so super, hence the need yo keep playing and keep learning. The way the guy was waxing lyrical about it, I'm assuming no-one else has an algorithm anywhere near capable of copying th
  • by xevioso ( 598654 )

    What would be amusing would be to get to the point where a person could generate their own set, print them out with the approximate levels of rarity necessay for a set, and then have a draft with the random cards.

    Who wouldn't like a card with MointainSpoink and Tromple?

  • It would be nice to see the same level of logic applied to scenario generators for games with large, persistent worlds. Skyrim and GTA games are what come to mind soonest for me. It would really pump up replay value. Probably neither of those games really needs it, but it would be a nice feature for a competitor.

  • There are a few interesting ways AI can be applied to the game of Magic the Gathering - this is one, and it's cool to see this guy's project, and the results.

    I've done a little bit of thinking in this area, but more in the area of making an AI play with the goal of searching for new competitive deck lists.

    I posted on my blog about my approach a while back: Getting Your Computer to Make an Awesome New Magic the Gathering Deck [hackshop.com]

    Another interesting tack would be to see if you can write an agent to create single

  • When I was first learning to code a long time ago I created a D&D character creator program. I thought a cool feature would be having the ability to randomly generate a character including attributes, class, proficiencies, spells, traits - literally everything. It was a lot of fun to start a new campaign and have each player use a randomly generated character. It really took you out of your comfort zone to play a ranger that was a dual wielding specialist with daggers or a wizard with pathfinding skil
  • by penguinoid ( 724646 ) on Friday June 12, 2015 @02:47PM (#49899839) Homepage Journal

    His next research project is to make an AI that defends against copyright infringement lawsuits.

  • Huh. I wonder if you could do anything interesting by linking the card database to the various databases of card sellers -- price, stock and sales rate, number of editions in which the card appears.

  • I noticed that the network, now more fully trained, could generate meaningful, novel cards. However, it also had a knack for generating profoundly useless cards. Here are a few snippets from the output:

    * When $THIS enters the battlefield, each creature you control loses trample until end of turn.

    Not a bonus, but plenty of creatures have slightly negative effects if they cost less to summon than their positive traits might suggest.

    * Whenever another creature enters the battlefield, you may tap two untapped Mountains you control.

    Weird, but if you're prevented from tapping mana sources for some reason...

    * 3, : Add 2 to your mana pool.

    Useful if you're tricked into a large mana-burn situation. It effectively reduces all mana-burn down to 1.

    * Legendary creatures can't attack unless its controller pays 2 for each Zombie you control.

    Oddly specific, but not useless.

    • I noticed that the network, now more fully trained, could generate meaningful, novel cards. However, it also had a knack for generating profoundly useless cards. Here are a few snippets from the output: * When $THIS enters the battlefield, each creature you control loses trample until end of turn.

      Not a bonus, but plenty of creatures have slightly negative effects if they cost less to summon than their positive traits might suggest.

      True, but having your own creatures lose trample is unusual as a drawback to a card, it's a very very situational penalty.

      * Whenever another creature enters the battlefield, you may tap two untapped Mountains you control.

      Weird, but if you're prevented from tapping mana sources for some reason...

      Due to a combination of how the stack works, and the majority of mana sources having the timing priority it's pretty much impossible to prevent a player from tapping a mana source for mana. The best you can really do is force the timing of it so that it's not useful for your opponent to tap a source for mana, for example if you use a spell or effect that would cause an opponent's land t

If all else fails, lower your standards.

Working...