Solved conversations

I was hooked on Mühle from the moment I started playing it. The board game, sometimes called Nine Men’s Morris, is one of the oldest known to man. It can be found at archeological sites, on the sides of ancient buildings, and at your local Wal-Mart. I personally discovered it when I visited my friend Attila in Germany last winter.

The first time we played it, I told him not to go easy on me. But after a few sore losses, I opted for a different approach. I decided the best way to learn the ropes would be to listen to him think out loud about each move before he made it. We played a few games like this, considering the best possible moves that he or I could make on each turn, until we came to a realization. When we put our heads together, neither of us made any mistakes. The specifics of the game aren’t important here. But by eliminating mistakes, we’d ensured that neither of us would win. Perfect play resulted in a draw.

Later that night, I did a little poking around on the web and discovered that Mühle is what’s called a “solved game.” It’s just one of a group of solved games that includes Connect Four, Tic-Tac-Toe, Checkers, Mancala, and Chopsticks. To solve a game is to correctly predict the outcome of the game from any position, assuming that both players play perfectly. Given a configuration of pieces on a Checkers board, for example, a computer can predict which player will win (if any) if both make the best possible move on each turn.

The concept of a solved game is a fairly recent one. Though Tic-Tac-Toe players have known for years that the game always ends in a tie if both players play perfectly, it’s not so easy to predict the outcome of more complex games. Until the 1980s, computer scientists and game theorists simply didn’t have the computing power to make those kinds of predictions.

How do games get solved? One of the more common techniques is called retrograde analysis. The computer builds a database of all possible terminal positions (i.e. all the positions that indicate the game is over, like checkmate in chess). The computer then constructs another database that contains all possible sequences of moves that could have led to the terminal position. Then, for any given position, it can determine the shortest sequence of moves to the optimal outcome.1

There’s a constraint, however, on what kind of games can be solved. In his book The Signal and The Noise, Nate Silver explains why IBM was able to program a computer to play chess:

In chess, we have both complete knowledge of the governing rules and perfect information—here are a finite number of chess pieces, and they’re right in plain sight.

When chance, negotiation or imperfect information are an important part of a game’s outcome, it’s impossible for a computer to solve the game. This is why Monopoly or Scrabble will never be solved. Even with complete knowledge of the governing rules, there will always be an element of uncertainty. Irrational trades or random distribution of letter tiles prevent a computer from having access to all the necessary information.

Conversations as Games

In everyday life, there are a lot of conversations that proceed in quite a predictable manner—more like Tic-Tac-Toe than Monopoly. When someone asks, “How are you?” you’re expected to respond with, “Fine” or, “Good.” Mention that somebody pronounces a word strangely, and the conversation will undoubtedly devolve into a comparison of accents: “melk” vs. “milk”, “roof” vs. “ruff”, etc. The conversation at a college party follows an equally predictable template; most people limit themselves to a very specific set of topics: how hot the room is, how drunk their friend is, how far away their next destination is. And post-game interviews with athletes almost always include talk of how they “brought their best to the court” and can only “focus on one game at a time.”

These are the kinds of conversations that feel more like scripted dialogues than interactions between complicated human beings. They’re “solved” conversations, like the games mentioned earlier. Each sentence is like a move in a game. And given the starting conditions, it’s easy to predict the outcome.

Solved conversations take the creativity out of interaction. They’re like the perfect players in a computer model, following an algorithm neatly mapped out for them by their culture. Conversation about the weather or a crazy weekend party may be boring, but they don’t seem to do any real harm. What’s dangerous are solved conversations about things that really matter.

When people identify deeply with a political party or a particular philosophy, it’s not difficult to predict the outcome of any given conversation. This is actually the underlying assumption of all debates on cable TV. Imagine what would happen if the conversations between panelists on Fox/CNN/MSNBC were not solved. One talk show guest would eventually say to the other, “You know what, I agree with you,” and then there would be a few moments of awkward silence and the host would cut to commercial break early. Given knowledge of the guests’ political beliefs, TV producers can pretty accurately predict the outcome of a conversation (namely, that neither side will change their opinions). These conversations are solved. Although they make for good entertainment, they stunt intellectual growth.

Of course, these sorts of back-and-forths don’t only happen on TV. They happen in debates about religion, philosophy, and politics in every day life. Conversations about controversial topics are liable to go in circles forever.

What does it say about our mental functioning during these conversations if they can be solved? Let’s return for a moment to the conditions that are required in order for a computer to be able to solve a game: all elements of chance, negotiation, and uncertainty must be removed. If our conversations can be solved, we’ve created an environment that’s as simple as a board game. We’ve dumbed ourselves down and removed all uncertainty and negotiation, making the outcome predictable. This is the opposite of what you’d expect in an intellectually satisfying conversation—an exchange of ideas where anything can happen.

When we treat conversations like scripted dialogues, the magic is gone along with all of the characteristics that make us human rather than machine. The beauty of a good conversation lies in its unpredictability. So let’s stop reading off scripts and un-solve our conversations.


  1. Van den Herik, H. Jaap, Jos WHM Uiterwijk, and Jack Van Rijswijck. “Games solved: Now and in the future.” Artificial Intelligence 134.1 (2002): 277-311.