AI and the Art of Go
Today, a once-distant milestone arrived much sooner than anyone predicted: AlphaGo has defeated Lee Sedol, and computers have conquered Go.
The world’s oldest and most intricate board game now joins chess, Jeopardy, and the host of other pursuits that, once exclusive to humans, have become the province of ever-more-sophisticated artificial intelligence. But this time is different: whereas previous conquests could be explained away as “brute force,” Go required the development of true intuition. Deep Blue’s chess differed from the human kind as dramatically as airplanes do from birds. In contrast, AlphaGo learned Go much as a human would: through observation, experimentation, self-improvement, the organic development of heuristics, and, ultimately, a keen feeling for master play. That feeling may be expressed in the form of neural nets and floating-point numbers, but it is feeling nonetheless, for what do you call a sense that tames the near-infinity of possible variations into one move, if not “intuition”?
As expected, the advent of AlphaGo has sent shockwaves throughout Asia and the world. Many have recognized it as a historic milestone in artificial intelligence, whose elements will prove valuable in technology, medicine, and a long list of other practical applications. But AlphaGo’s triumph has also provoked a special sadness. This is no ordinary game that has fallen; it is one of humanity’s oldest, most profound, and most beautiful intellectual arts. Edward Lasker once claimed that Go is “so elegant, organic and rigorously logical that, if intelligent life forms exist elsewhere in the universe, they almost certainly play Go.” Throughout its 3000-year history, the game has been considered a hallmark of culture (one of the Four Arts of ancient China), an expression of human genius, and a mirror to the soul. One must wonder if a computer’s victory in such a game tarnishes that mirror. And if the mirror is tarnished, isn’t our image of the human soul dimmed as well?
I’ve wrestled with these thoughts myself, particularly as some of my closest friends have shared this perspective. To me, and to many of the 40 million other players in the world, Go represents more than a game. It is a strangely bewitching thing, these patterns of wood, slate, and shell, this system of such simple design and such impossible complexity. I have never found anything so intellectually captivating, and doubt I ever will. Mathematics itself must be benevolent if the wonderful patterns of the Go board can arise merely, in conformance with a minimal set of rules, from placing one stone down after another.
And now to the crux of the matter: as everyone who has felt this way knows, there is something mysterious and nearly magic about this process. That sense of mystery has always been deeply cherished among Go players. It is the reason we were so proud that Go, not merely a game but an art, stood unconquered among games of strategy. But now a program reigns supreme, not a human mind, and in the presence of science there can be no mystery.
That sense of loss: it feels as if the art of Go is gone.
But must we write the epitaph of Go, now that it has been subdued? Has the game that endured for millennia at last entered its twilight?
What will be the fate of Go?
To answer that question requires a more precise definition, because Go is not one game only, but three: Go the science, Go the mind sport, and Go the art. Each of these will react very differently to AlphaGo’s victory.
Go the science is essentially finished, because the interesting questions about the game are essentially solved. The first Go program was an ALGOL script written in 1968, and since that time we have progressed through three epochs of research. The first, until 2007, devoted its effort to hard-coded expert knowledge and heuristics. The second, until 2014, built on the Monte Carlo tree search techniques pioneered by Coulom, Gelly, Silver, and others. The third, driven by DeepMind, lasted only two years but comprised the greatest leaps in strength, delving into deep neural networks and reinforcement learning. The interesting questions of Go were: Can computers match humans? Can they do it by brute force, or must they develop intuition? And is it possible to create a strong value function for the game? The answers were yes, intuition, and definitely. But science succeeds best when it renders further study unnecessary. For the research community, the existence of AlphaGo is an unmitigated triumph, in that it has done just that. If our appetite for research is not yet sated, we will need to devise an entirely new set of questions.
The sport of Go will change rapidly, in much the same way as the sport of chess did. After Kasparov was vanquished by Deep Blue, chess computers quickly matched humans at smaller scales, first on desktops and then even on phones. No longer were stables of grandmasters necessary to prepare a champion for the next game, as grandmasters, aspiring professionals, and amateurs alike gained access to superhuman-level instruction and analysis. The theory of the game experienced a renaissance as computers introduced a flurry of new ideas. Cheating also ran rampant for a while, then was quickly curtailed as the same computers that had enabled cheating prevented it through move-by-move Elo analysis. Amateur spectators gained a better understanding of masters’ games, with live engine-based analysis indicating who had the advantage and what alternative lines of play they could have chosen.
Whether AlphaGo is involved in these developments or suffers the sad fate of Deep Blue, shut away forever by IBM after its last match, the Go world will see all these changes take place over the next several years. And though the sport of chess looks different now, with computers integrated into every facet of competitive play, the chess community has not suffered unduly. Magnus Carlsen is a household name in Norway, famous enough even to create his own clothing line, though he has never beaten a top computer and never will. The name of Lee Sedol will remain dear to Koreans, regardless of his defeat in this series, and his place in history as one of the greatest players of his era is secure.
That leaves only the art of Go. In this perspective, which finds expression throughout East Asian history, Go is not a contest of victory and defeat, but a poetic and semi-mystical game that both captivates and reflects the spirit. One famous legend, Lan Ke (“Rotten Axe”), finds a young man lost in the forest where he has gone to cut wood. He stumbles on two strange old men playing Go, one of whom gives him a date to eat. He sits down to watch their game and, rapt with attention, falls into a trance. When he finally arises, he realizes that so much time has passed that the handle of his axe has rotted away.
The art of Go also implies a certain insight into a player’s personality. The Go board is large enough to accommodate every type of attitude: aggression, calm, fear, desire, greed, hope, rebelliousness, and lust for power are only a few of the traits that can be directly recognized in any given move. In Edo-era Japan, where Go reached its cultural height, players would joke that a true master could look at a game and determine when the student was calm, when they were confident, when they were desperate, and when the maid came by with tea.
And in all three of the major Go-playing countries (China, Korea, and Japan), the concept of the “divine move” still commands the imagination: that is, a move so profound that only God could play it. The idea is ancient, but it survives in many forms today. In fact, a movie called The Divine Move came out in Korea in 2014, to enormous commercial success. The anime Hikaru no Go, which has inspired a large surge in popularity for Go over the past fifteen years, also makes frequent references to “Kami no Itte”. It seems that Go players simply cannot help but inject an element of spirituality into the game.
All of this sentiment is tied to one perspective: that not only is Go a human game, it is specially human. It is part of what makes humans unique, the ability to fathom something so mysterious so deeply, and synthesize all this information into one right move. It belongs to that part of human character that approaches the divine.
And it is that sacredness that the victory of a computer has taken away. For that is the essence of science: replacing mystery with understanding.
Many articles on AlphaGo have discussed the future of Go, and most of them use a physical analogy. Did people cease to run when the car was invented? Of course not, they say, and so it will be with Go. But if Go is an art, this analogy remains deeply unsatisfying. The car does not run like we run. But AlphaGo plays like we play, only better. In the words of Lee Sedol’s rival Gu Li, commenting after the second match, “it played some moves that only God could have played.” AlphaGo, it seems, has proven more divine than we have.
Yet I do not believe that this match ends the art of Go. I believe I have a better analogy.
Astronomy has always been a deeply inspiring subject: what lies above us is perhaps the only thing as astounding as what lies within us. And astronomy, too, has undergone a transformation in knowledge, but its moment came a very long time ago. We once thought the sun revolved around the Earth; that is, that the universe itself was anthropocentric. Copernicus proved that theory false, and suffered for it, but ultimately the universe became no less marvelous for his discovery. Instead it grew more so, as we looked first beyond the bounds of our own planet, then of our solar system, then of our galaxy. Our horizons did not contract, but expanded, when the universe in our minds stopped revolving around us.
Now we have discovered that Go does not revolve around us, either. We may soon find that this is true of other pursuits we hold dear, such as music or writing. But Go itself is no less sublime, no less fantastic, for its submission, and nor would any of these others be. A game of Go holds as much interest for me today as it did yesterday. In fact, it holds more, as the creative and counterintuitive moves in this very match illustrate how much uncharted territory remains. With that clarity comes a chance for understanding, an opportunity to map that territory. What is the true komi? What is the best variation of the avalanche? What is the true strength of any given move? And who was the strongest player ever to have lived? If computers had never mastered Go, we would forever continue to answer these questions in a limited and temporary way. But now, one day, we will know. And the emergence of such clarity offers an opportunity to unite the Go world in a way that has never happened in its millennia of existence.
The moment may be bittersweet, but the art of Go has not died today. Today, the art of Go is reborn.