The perspective-expanding magic of Move 37

and now I want to learn a new board game!

The more I see this move, I feel something changed. Maybe, for humans, we think it's bad, but for AlphaGo, why not?

— Fan Huei, 2p Go professional

Today, I want to talk about Move 37 and what it means for how we see the world, how we see ourselves, and how we relate to that which we create.

First, some context:

On Wednesday, a dear friend sent me a link to a 2016 documentary called AlphaGo. I take Andrew’s recommendations to heart, and I’m thankful I do because I haven’t stopped thinking about the film in the days since.

AlphaGo is basically an artificially intelligent robot created by DeepMind that is programmed to excel at the game of Go. Games shouldn’t be too hard for A.I. these days (or at least back in 2016), right? After all, the Deep Blue chess computer beat the world chess champion way back in 1997. Well, Go is not quite chess—rather, the simplicity of the game creates conditions that are very much dependent on what we might call human intuition. As Demis Hassabis, co-founder and CEO of DeepMind (and just a wonderfully brilliant and curious individual) says early in the film:

Go is incredibly challenging for computers to tackle because compared to, say, chess, the number of possible moves in a position is much larger. In chess, it's about 20. In Go, it's about 200. And the number of possible configurations of the board is more than the number of atoms in the universe. So even if you took all the computers in the world, and ran them for a million years, that wouldn't be enough compute power to calculate all the possible variation.

Well then! This is a significant challenge for DeepMind—whose long-term aim is to solve intelligence, advance science, and benefit humanity—and the world of artificial intelligence at large. The film follows the AlphaGo team of engineers as they build and improve their A.I. with the goal of having it beat the very best human players, a feat that many thought to be decades away when it was first conceived. Hassabis, who was the second-ranked chess player in the world as a youth, sees games as a wonderful structure in which to test A.I. systems and solve problems.

Professional Go players are ranked by their dan classification (one dan, or 1p, all the way up to 9p). First, AlphaGo takes on Fan Huei, a three-time European Go champion with a 2p ranking (this may seem low, but amateurs have to work their way past seven different classifications before getting a 1p ranking). AlphaGo wins the five-game match as convincingly as possible, 5-0. It isn’t even close. Huei, one of my favorite characters in the film, is pretty embarrassed at first—an emotion that he confronts admirably, choosing to embrace the curious mystery behind a project like AlphaGo and later joining the DeepMind team to help point out programmatic weaknesses.

Huei is a formidable test, but he’s not the test. A few months later, the AlphaGo team travels to South Korea to pit their machine against the Roger Federer of Go, Lee Sedol. Sedol is an 18-time world champion and has a 9p ranking. If there’s a person on Earth to see how good AlphaGo really is, it’s him.

The five-game match, much to the shock of Sedol and the millions of people watching around the world—and, to a degree, the AlphaGo team—starts much like the match with Huei. AlphaGo wins the first game, and then the second one. That in itself is kind of wild to think about. Watching the grandmaster of Go, in real-time, as it dawns on him that he’s been outwitted, and that he has no choice but to resign, was a strangely emotional moment for me as a viewer. But it’s something particular that happened in the second match that has captured my mind in the days since. Something that prompted Sedol to remark, “Yesterday I was surprised, but today I am quite speechless.” That something was Move 37.

Here is the moment in the film when Move 37 is generated by AlphaGo’s algorithm and played by its proxy human, Aja:

One of the live stream commentators giving analysis (a 9p Go player himself) hesitates when imitating the move for those watching the live stream board because of how surprised he is. “I wasn't expecting that. I don't really know if it's a good or bad move at this point,” he fumbles. Almost everyone thinks it’s a mistake, but they don’t really know why. Fan Huei, with his trademark enthusiasm, puts it best:

When I see this move, for me, it's just a big shock. What? Normally, humans, we never play this one because it's bad. It's just bad. We don't know why. It's bad!

In other words, the move seems to be “bad” because, over thousands of years of playing Go and developing insights/strategies, the human Go-playing collective has deemed it so. This probably wasn’t deliberate, but rather a slow accumulation of shared wisdom around “this is how good/bad moves look,” which unavoidably places some sort of lens through which you or I or anyone else might look when first learning and developing our own strategies for playing Go. AlphaGo’s algorithm seems to give some sort of hint as to why this is an unthinkable move by our species’ standards: it gives Move 37 about a 1/10,000 chance of being played by a human. David Silver, engineer and leader of the AlphaGo project, remarks that the A.I. went “beyond its human guide” in this instance.

As it turns out, the move was mind-blowingly brilliant, successfully connecting other pieces on the board in a way that was apparent to pretty much nobody but AlphaGo—even DeepMind’s engineers, who built the algorithm, thought it was some sort of error. It was anything but an error, instead revealing an unconventional strategy that, as it became apparent, left viewers ranging from complete novice (me) to 9p wizards feeling something in common: awe. Sedol has to get up and leave the room to ponder what AlphaGo might be trying to do with such a move, and it’s only a matter of time before he resigns.

Reflecting on Move 37, Sodol conveys how radically his perspective on AlphaGo, and the game of Go as a whole (remember, this is the most accomplished player on Earth) shifted in a single moment:

I thought AlphaGo was based on probability calculation and that it was merely a machine. But when I saw this move, I changed my mind. Surely, AlphaGo is creative. This move was really creative and beautiful.

and a little later in the film:

What surprised me the most was that AlphaGo showed us that moves humans may have thought are creative, were actually conventional. I think this will bring a new paradigm to Go.

Take a moment to think about how one decision made by a computer (built by humans) completely shifted how people around the world define “good” and “bad.” In this case, it was within the realm of Go. In which other realms are we in dire need of a perceptual reset—a reframing that may seem inaccessible simply because of the stories we tell ourselves? After thousands of years of playing and refining a game like Go through the lens of human understanding, why was Move 37 so unfathomable that it was initially deemed an error? What does the high likelihood that Move 37 may have never been played without the materialization of AlphaGo mean for our collective cognitive limitations?

In other words, what else have we become so sure about (“that’s just how it goes”) that we haven’t taken the time to go back to the very beginning and wonder: is there a better way to go about doing this? And, by starting with such a blank slate, seemingly indifferent to emotion or “how things are”, how might artificial intelligence, point out our blind spots and bring about new paradigms in other subject areas, for better or for worse?

Rick Rubin has some thoughts.

Rubin—legendary music producer and sauna enthusiast—recently stopped by the Rich Roll Podcast (the RRP shoutout train continues to gain steam) to discuss his new book, The Creative Act: A Way of Being, which I’m very excited to pick up. And it turns out that’s why Andrew sent me AlphaGo in the first place because Rick and Rich (a quadruple ‘R’!) got to chatting about it. And I loved their insights, which start here:

I was particularly struck by Rick’s recounting of his experience seeing the film. Upon seeing Move 37, he found himself crying and didn’t know why. After reflecting, he realized it wasn’t because machine had convincingly beaten man for the second game in a row—something that caused a profound ripple of sorrow amongst others in the film—but because of something else, something sort of liberating and beautiful:

The reason the computer won was because the computer didn’t know more than the human. The computer knew less than the human.

In other words, because the computer embodied the principle of the Beginner’s Mind and didn’t know the customs/morals attached to the history of Go and “how it should be played,” it actually created a better outcome from knowing less. Rick’s insight here is that we don’t always necessarily need to know more. Rather, we might make a massive jump in creative progress and solve some seemingly complex problems simply by looking at the same thing or problem in a different way.

A lesson we can take from this is that when something, anything, happens a certain way, that might be the way—but what else could it be? Thanks to the wonders of language and human cooperation, etc., sometimes the way of conventional wisdom is indeed the way. But only sometimes! Things are not always—in fact, almost never—as they seem. What kinds of magic can happen when we take a moment to consider other possibilities, to start from first principles?

In committing to open-mindedness, we gain much. We gain the ability to work with machines like AlphaGo when they deliver a healthy dose of humility and build other tools that can help us check our biases and reach new planes of understanding one another and the world.¹ We gain the ability to drop old stories that have cemented themselves in our head (“Well, that’s just the way I am.” “Those people just don’t get it, and they never will”) and consider new narratives. We gain the ability to experience more awe in our day-to-day living, choosing to see not knowing—the mystery of everyday life—as a source of great joy and curiosity, rather than anxiety, and we become open to inspiration from anything and everything around us.

I will close with another great quote from project leader David Silver, who displays a deep understanding of the synergistic, mind-expanding potential of technologies like AlphaGo. This all comes with a giant caveat of all the things that could go horribly wrong—both with and without intent—if they aren’t approached with caution, a commitment to transparency, and so much else. These tools have very serious implications beyond the scope of games, and I’m eager to hear more from Hassabis about his approach (I’ve got this podcast queue’d up). But for now, I am very much enjoying the afterglow of awe and wonder that came from watching AlphaGo, and from having my mind expanded by Move 37.

There are so many possible application domains where creativity, in a different dimension to what humans could do, could be immensely valuable to us. And I’d just love to have more of those moments where we look back and say, "Yeah, that was just like move 37. Something beautiful occurred there." 

1. This is demonstrated in the case of Fan Huei, who, as I mentioned, went to work with DeepMind after getting beaten by AlphaGo. In choosing the path of humility and curiosity, which led to a deeper understanding of the computer’s strategizing process, Huei himself improved as a Go player. To quote from a fun Wired article:

The experience has, quite literally, changed the way he views the game. When he first played the Google machine, he was ranked 633rd in the world. Now, he is up into the 300s. In the months since October, AlphaGo has taught him, a human, to be a better player. He sees things he didn't see before. And that makes him happy. "So beautiful," he says. "So beautiful."