What AI Teaches Us About Magic

In a rapidly evolving, technology-based culture, perhaps the ultimate demonstration of human advancement is the concept of artificial intelligence. The very idea that human beings can create machines that could independently think for themselves and solve problems is both exciting and terrifying at the same time. It is the definition of the sublime.

Play Like R.U.R.

Whether you are the type of person who looks at a moment in time such as this with excitement (so many possibilities!) or utter terror (did you see the Terminator movies?) it is pretty safe to say that the fact that we live in an age where the possibility is quickly approaching reality rather than science fiction fantasy is astounding.

I recently began following an interesting story that relates to both gaming and artificial intelligence: AlphaGo.

For those of you unfamiliar with the program, allow me to quickly catch you up to speed on the basics. AlphaGo is a program developed by Google DeepMind to play the Chinese board game Go.

Back in 1997 when Deep Blue defeated Gary Kasperov at chess, it was believed that the same result would not be possible for a Go-playing AI. For comparison’s sake, in any given game of chess, the average number of possible moves averages roughly 20 whereas in Go, the average number of possible plays ranges around 200! The decision trees are exponentially more complex and complicated.

Another interesting factoid I’ve encountered multiple times in reading about Go is that the elite, professional players describe intuition as a major contributing factor when determining which play to make among so many possible options. Since there are so many possible moves, an intrinsic knowledge of which types of plays you should make is important. How could an AI program possibly learn, process, and replicate this element of human intuition to actually defeat top-level players?

Well, as it turns out, perhaps human intuition isn’t the most important and unreplicable element when it comes to gaming. The AlphaGo program has now defeated two renowned master champions without losing a game! Oh snap—robot got game.

The Intelligence of Options

So, we are roughly halfway through the article and I haven’t even mentioned MTG yet. I’m just talking about non-affinity robots and an ancient Chinese board game that most people don’t even know how to play!

Here is where things get interesting and become very relevant to a discussion of MTG strategy and tactics: AlphaGo essentially taught itself to play Go and developed its own strategy and approach to play a game that operates, at least according to the testimony of professional players, on a highly intuitive level. Advanced analysis of AlphaGo’s strategy and play reveals one interesting fact:

What AlphaGo valued most highly was making plays that created or opened up more options on the following turns.

It didn’t value specific known strategies, it didn’t try to play to any intrinsic spot on the board, it didn’t try to execute any specific endgame—the program assessed that having more options and creating opportunities were the most important factors in this particular infinitely complex game.

Alex Wissner-Gross has a mindblowing video “A New Equation for Intelligence” in which he discusses the nature of AI, and intelligence in general, and postulates that it can be expressed as an equation:

F = T ∇ Sτ

He says, “Intelligence is a force that acts so as to maximize future freedom of action, or keep options open with some strength with the amount, the diversity of possible accessible futures on some future time horizon. In short, intelligence doesn’t like to get trapped.”

So, for you guys and gals out there who want to play the field and are being pressured by somebody to settle down, you can tell them that you are applying the universal equation for intelligence to your romantic life and are going to continue to play the field like an unstoppable Go master!

Anyway, take a second to let the idea behind that equation settle in because you already know where I’m going with this in an article about Magic strategy.

The Importance of Options in Magic

If you think about the kinds of decks that have traditionally done well throughout the history of the game, you’ll find a direct correlation between decks that maximize options and winning. It seems obvious enough when framed in the context of this conversation.

Decks like Faeries, Jund, and Control Slaver all immediate strike me as dominant “best decks” in their respective formats where the deck construction is flexible and leaves options open to the level of an art form. Flexible answers, instant-speed interactions, and cards that can execute or attack various angles of game play are accentuated in these decks.

Another interesting angle that I’d like to discuss with regard to the intelligence equation and Magic is the idea of “taking away” an opponent’s options. While one theme of “good decks” is certainly to create more options, another angle is to directly impede your opponent’s options.

I like using Vintage as a vessel for talking about these ideas because by definition, Vintage is like Go in the sense that on any given turn, it allows players the most options by virtue of having access to a larger and more powerful card pool.

In Vintage, MUD Prison decks counter-intuitively work to eliminate an opponent’s options right from the get-go. They constrain an opponent’s mana in such a way that on any given turn, the opponent might have one or even zero viable options. If you consider that “intelligence” is creating more options, and that access to options is valuable, then inhibiting options is a fantastic strategy! People tend to talk about Shops, Legacy MUD, or Modern Colorless Eldrazi as decks that leverage advantages in mana. While that is true, that leverage is actually in service of the larger goal of impeding options and thus having more options than the opponent.

Card advantage is also a means of leveraging options over an opponent. If you have five cards and they only have one card, then you have significantly more options than your opponent. But you know that “card advantage” isn’t as important as options. Have you ever died with six cards in your hand? Yeah, me too. Ever lost to Dredge?

Cards don’t directly equate to options if they can’t be played or don’t matter. If you are at 1 life and your hellbent opponent topdecks a Lightning Bolt and you don’t have a counterspell or life gain spell, then all those cards don’t create viable options.

The Top 5 Cards AI Would Love to Play

I believe that AI will eventually possess the capacity to feel. So let’s end this on a fun note of the Top 5 cards that an MTG playing AI would love to cast!

I have this funny image in my mind of a Magic playing computer whining about how it got so unlucky that its opponent topdecked a Reality Smasher, or never drew a third land as it gets promptly crushed at FNM.

AlphaGo is a pretty sweet name for an unbeatable Go-playing robot, so I feel like I should name my imagined, sci-fi MTG playing computer too: Skynetdeck, AlphaDrawGo, AlphaCawGo, AlphaBetaUnlimited, DeepMonoBlue, or DeepUWControl?

I also considered naming it Owen Turtenwald, but apparently that name is already taken by another unstoppable MTG supercomputer.

#5. Sensei’s Divining Top

If our robot friends want to maximize their ability to create new options, then I’m fairly certain that Sensei’s Divining Top is a card they will be looking to play with. The ability to continuously see more cards and manipulate draw steps seems like the epitome of leaving options open.

I can just imagine the eye roll of the super-computer when you also show it Counterbalance and Terminus

#4. Demonic Tutor

Tutor cards are great because they provide you with the option of any card left in your library. “It could be anything… Even a boat… Lois, you know how bad we’ve always wanted a boat!”

Having access to not only the cards in your hand or potentially the graveyard, but also any card in your library to strategically interact with is awesome. There is a reason cards like this are banned outside of every format except Vintage.

As an aside, I’d personally like to think that an MTG AI with the capacity to feel would really enjoy and appreciate the Vintage format.

#3. Arcbound Ravager

And not just on the bias/preference to play with robots! Ravager is a great card because from the moment it resolves, your options increase exponentially, at instant speed, for free.

One of the reasons I’ve always believed that Affinity was a great deck is because this card allows you to suddenly do so many different things with your board. Also, I have always maintained that Affinity was one of the most difficult decks to pilot “well” even when the common perception was that it was a “scrub” deck two or three years ago. Yes, the deck can win in the hands of way below average players on the strength of certain types of draws, but the deck also provides skilled players with tons of options on any given turn.

#2. Counterspell

Not specifically Counterspell, but I think that the flexibility of permission is something that would interest an MTG AI. Permission gives you flexible answers to almost anything, including other counterspells!

I have a strong intuition that MTG AI would either play a prison-style MUD deck or a blue-based control deck since those two options create the most options or take away the most options with a great deal of efficiency.

#1. Tinker

It tutors, which gives you flexibility, and is robot themed? I can’t imagine an MTG playing AI that wouldn’t be excited to cast a card like this!

Can you even imagine the level of creating options that Tinkering for a Top would create for the machines? Makes a strong statement: “Bite my shiny metal buns.”

Imagining what cards an MTG robot would enjoy playing is all fun and games, but I think the lesson we can take away from AlphaGo is very real. The strategic element of Magic gameplay does reward this idea of “keeping options open” and “creating or taking away more options.”

I haven’t read everything on AlphaGo but I am following the topic with some interest. As always, feel free to hit me up on FB and Twitter, or drop a comment below. I’m always interested in continuing and following up on the discussion with you all.

Until next time. Hasta la vista, baby.


Scroll to Top