Talcos, a PhD candidate researcher in computer science was investigating the use of deep neural networks for problem solving tasks and classification.
For example, if you force the network to read Shakespeare over and over and eventually it can write its own texts in the style of Shakespeare. -Talcos
Then he thought “what if it could do the same for Magic cards?” He fed the RNN all the Magic cards ever created and it started reading them over and over and noticing patterns. In the beginning, it spewed out nonsensical cards with wacky effects and names. Though, keep in mind that the RNN has no prior knowledge of english, Magic, or anything for that matter.
A few cards that came out early in the training:
Artifact Creature – Kavu Shaman
Morph (You may cast this card from your graveyard to the battlefield, you may pay . If an enchantment card, then put the rest of target creature card from your graveyard for its flashback cost. If exile is you sacrifice it unless you pay . If you do, put a 3/1 green Soldier creature token onto the battlefield. Put it into your graveyard.)
Creature – Dragon
(Yes… That is Tromple and Mountainspalk 🙂 no typo)
Instant – Arcane
Exile target creature you control.
A few things before we go through more cards. The names are pure nonsense. It apparently commonly churns out new keywords like Tromple and Mountainspalk. It is cool that it knew to put reminder text behind the Morph on the first card, but the reminder text is gibberish. It also put B/B on the morph cost meaning that you can pay 1 black or 1 black? Interesting.
When he let it run overnight, he found it in the morning churning out almost meaningful and playable cards:
Light of the Bild
Creature – Spirit
Whenever Light of the Bild blocks, you may put a 1/1 green Angel creature token with flying onto the battlefield.
Artifact – Equipment
Equipped creature has fuseback. (Fuse and Flashback fused together)
Horror deals 3 damage to target creature or player.
The cards above are meaningful and some of them are playable. It still has some problems, like fuseback, but much better. Below is a gallery showing some of the cards.
Very interesting work Talcos is doing. To see Talcos’s original MTGSalvation post on it click here. To view Robo Rosewater’s (Named After Mark Rosewater) Twitter Account that posts RNN generated cards, click here. The source for the photos is the Nerdist, and the Nerdist is where I originally read about this. Thanks The Nerdist! Thanks for viewing my post, and I hope you enjoyed!