BY PATRICK HOUSE
Midway through the first of five recent matches between Lee Sedol, a top-ranked professional Go player, and AlphaGo, a computer program conceived by Google DeepMind, an odd thing happened: Lee’s jaw dropped, hanging open for a nigh-cartoonish twenty seconds, and then he laughed. AlphaGo had just mounted an aggressive, and evidently unexpected, attack. The moment was reminiscent of a famous episode in Go history, when Honinbo Shusaku, a future legend of the game, squared off against Inoue Genan Inseki, an older and more experienced player, in 1846. The story goes that a spectator—a local doctor who knew little of Go—correctly guessed that the seventeen-year-old Shusaku was beating Inseki. Asked how he knew, the doctor responded that, after an earlier move, Inseki’s ears had flushed red, a clear indication of surprise.
Earlier today, AlphaGo won the final game in its tournament against Lee, for an over-all record of 4–1. As recently as 2014, it was thought that humans would remain competitive at Go for at least a few more years, maybe longer, because the game’s nuances make it particularly hard for an artificial intelligence to grasp. Now that assessment appears to have been wrong. “We are out of the A.I. winter,” Eric Haseltine, a longtime technology researcher and a veteran of the National Security Agency and the Office of the Director of National Intelligence, told me. “If you look to the future, we are probably not even at the knee of the curve of what A.I. can do. We are at the beginning of the beginning.” Like other tools—electricity, for instance—A.I. will likely go through many iterations until, one day, we barely notice its presence. And so, too, will it be used for good and, just as likely, for not-so-good. “I don’t know of any tool that has evolved that hasn’t been used as a weapon,” Haseltine said.
The concern is of particular relevance to Go, which has long been considered a kind of war in miniature. In 2011, Henry Kissinger published the book “On China,” in which he argued that understanding the twenty-five-hundred-year-old game, known in Chinese as wei qi, is essential to understanding Chinese thinking and military strategy. Kissinger contended that if the proxy wars between the West and Russia in the twentieth century were chess-like, then twenty-first-century diplomacy between the United States and China might be played in Go terms. Critics of the war-as-game metaphor point out that the real world is, if a game at all, a stochastic and messy one. Nevertheless, David Lai, who grew up in China and is now a professor at the Strategic Studies Institute of the U.S. Army War College, told me that the argument is useful as an aid to understanding the language and thoughts of others. Last year, Lai noted, the Chinese foreign minister claimed that China was moving out of the bù jú, or opening stage, of great-power diplomacy. They were moving, in other words, on to the mid-game. “I bet many American leaders find those terms confusing,” Lai said. “Where did that come from? What does that mean? But if you know about wei qi play, you can appreciate what they mean by that.”
It is not at all unusual for heads of state to borrow linguistic terms from sports or other popular pastimes. Lai, though, is interested in the strategic entailments of such usage. He gave the example that each side’s Go pieces, called stones, are all physically the same, unlike the specialized and dynamic pieces of chess, where even a pawn can become queen. This means that the location of a stone relative to other stones, rather than its intrinsic or acquired abilities, is the source of its power. “Depending on where you put them, the stones’ influence varies greatly,” Lai said. “It is natural for a Go player to think in terms of influence and the radiance of that influence.” He gave as a real-world example China’s recent pushes into the contested South China Sea. Lai believes that when two nations meet in a diplomatic or military encounter, they tend to want to, on a simplified level, play the game they know best. “The United States plays Go by accident,” he said. “The Chinese play Go by design.”
Could some of the algorithms behind AlphaGo, which learned to predict human behavior after being fed data on millions of expert Go moves, have military or diplomatic applications?* And, more to the point, would the A.I. again leave humans in the dust? Isaac Asimov’s “Foundation,” published in 1951, introduced the idea that, given a sufficiently large population, the broad movements of history—like, say, the movements of a gas—become statistically predictable. But Haseltine argued that modelling games is less useful than being able to model the people playing them, and A.I. isn’t there yet. “Getting inside the head of key actors in the world and terrorists is the holy grail of intelligence,” he said. “If you ask any intelligence officer ultimately what business they are in, it is, ‘We tell stories that prevent surprise.’ ” Perhaps the alternative to a future foretold by learned machines, then, is one in which computers help us avoid the ear-reddening moments. “We tend to think of A.I. as a problem-solver,” Haseltine said. “What if the best use of A.I. is as a problem-finder?” Consider an issue like climate change. “We, as humans, as neurons, as selves, will never and can never figure it out,” he said. “But will we as a hybrid of A.I. and humans figure it out?”
During AlphaGo’s match with Lee, the program’s physical moves—it has no hands, after all—were played by Aja Huang, a member of the DeepMind team and a strong amateur Go player himself. (Huang, according to his Go ranking, would have about a 0.7-per-cent chance of beating Lee.) In many ways, Huang embodied a possible methodological future for A.I.: for a given set of problems, we will require A.I.’s help, but it will also require ours. Huang sat across from Lee, acting as a sort of A.I. conduit—the last mile—and it must have been odd to see what few people ever have from that perspective: a Lee Sedol resignation. A popular animated television show in Japan, “Hikaru no Go,” tells the story of a ghost of an ancient Go master that inhabits a boy’s body. It does so as a coexisting but separate personality, showing him good moves. According to the fictional timeline, the ghost’s last host was Honinbo Shusaku.
_________________________________________________________
Structure of the Lead
|
|
WHO
|
AlphaGo and Lee Sedol
|
WHEN
|
March, 2016
|
WHAT
|
|
WHY
|
|
WHERE
|
|
HOW
|
Keywords:
- conceive (v.) 構想出
- reminiscent (adj.) 回憶往事的;發人聯想的
- square off 擺好姿勢
- spectator (n.) 觀眾
- tournament (n.) 錦標賽
- nuance (n.) (色調,音調,意義,見解等的)細微差別
- assessment (n.) 評價;估計
- veteran (n.) 老兵;老手;富有經驗的人
- curve (n.) 曲線
- iteration (n.) 重複
- relevance (n.) 關聯;適宜;中肯
- metaphor (n.) 隱喻
- stochastic (adj.) 推測學的
- entailment (n.) 繼承人之限制
- pawn (v.) 抵押 (n.) 當岀物
- radiance (n.) 發光;光輝;輻射;欣喜的神色
- algorithms (n.) 互除法;演算法;規則系統
- amateur (n.) 業餘從事者 (adj.)業餘的
- conduit (n.) 導管