Reed Morgan Milewicz is a Ph. D candidate working in the iProgress lab at the University of Alabama. His fascinating (and viral) simple artificial intelligence program has been “learning” Magic: The Gathering cards, and has become capable of generating its own cards from patterns of other cards. How does it do it, though?
It’s a pretty simple concept, and one you can replicate, if you’re nuts enough to dedicate a massive amount of computing power to your AI. First you set up your AI with as much processing power, RAM, and storage as possible. Then program it with an appropriate AI. You can follow some great instructions on designing the appropriate neural net from hardcast_sixdrop.
The hardest part is inputting the thousands of Magic: The Gathering cards to “teach” the AI. hardcast_sixdrop also has a solution that’s accessible by the public via GitHub. It takes an astounding amount of time, but eventually the AI reverse engineers the cards, to break down the patterns and complex design rules.
Milewicz let his AI run for a while, learning cards, drawing conclusions, and designing cards. The project is ongoing, but his AI is capable of creating legitimate (if sometimes odd) cards. My favorites include:
- Duren of the Laens
Instant
Duren of the Laens deals X damage to target creature and you gain X life, where X is the number of creatures you control. - Plaran, the Kiss of Chaos
Legendary Enchantment Creature – God
Indestructible
As long as your devotion to white and blue is less than seven, Plaran, the Kiss of Chaos isn’t a creature.
Whenever an opponent casts an artifact spell, you may put a 1/1 white horror creature token onto the battlefield.
5/3
Duren of the Laens is a fantastic Instant-speed spell, and a rare example of the AI using an “x” cost ability properly. Plaran is a fantastic example of great mechanics that are technically great, if a bit “off” in flavor. A white horror summoned by a Green god card is not something you’d consider, for flavor reasons. Add the Indestructible, 5/3, and a converted mana cost of 3, and you have a seriously overpowered card. I’d play it!
The project is ongoing, and you can check out the design skeleton in a Google Doc. The community feedback is making an ever-growing powerhouse tool. Milewicz has started an entertaining project that draws geeks from all realms. If you try it for yourself, let us know! We’d love to see what kinds of cards your personal AI designs. Email Rory and we’ll share!
Duren could not be a mono-red card, as Red isn’t to gain life. The card would probably be Black if mono color, and more likely Golgari at 3BG.
I agree, to a point! I would like it best as a red/white card, like Warleader’s Helix. Instant speed, direct damage, correlated life-gain? Yes, please! Because of the “x” factor, I’d bump it to a 3RW, personally. Even fits the Boros model, my go-to dual-color deck style.