Finding Solace in Defeat by Artificial Intelligence
Fan Hui, the European Go champion, needed some fresh air. “I don’t understand myself anymore.”
Hui was the first professional Go player to face AlphaGo, Google’s artificial intelligence system and the title of a new documentary by Greg Kohs that debuted last week at the Tribeca Film Festival in New York. When Hui was invited to visit Google’s London office housing the DeepMind research group that developed AlphaGo, he was feeling confident. After all, as Hui puts it, “it is just a program.”
Hui had reason to be confident. Although AI has achieved impressive feats in the last few years, Go has been a longstanding and especially daunting challenge—the “pinnacle of board games” says Demis Hassabis, cofounder and CEO of Google DeepMind and a world-class game player himself. No program had ever defeated a human professional on a full-sized board, and when Hui played that afternoon in London, he wasn’t expecting a challenge.
But AlphaGo was no ordinary Go program. It was the product of innovative engineering, and the teamwork of a couple dozen scientists at DeepMind. Importantly, it also played a superhuman amount of Go, using that experience to train its deep neural networks and improve its play. By the time it faced Hui, AlphaGo had trained on 160,000 games recorded from top Go players, and then on 30 million more games it played against itself. As Hui found himself losing to AlphaGo, he knew that his world—and the world of professional Go—was about to change forever.
In a film that documents a technical achievement, I saw two human stories. There is the work of the DeepMind scientists and their impressive quest to reach this AI milestone. But there is also the story of Fan Hui and others who dedicate their lives to studying this game. Cinematically, the latter story shines. How does it feel to no longer be the best? If mastering Go “requires human intuition,” what is it like to have a piece of one’s humanity challenged?
As AI gradually changes our daily lives, we will all need to grapple with these questions. For Fan Hui, in one moment he is the European Go champion. In the next he is hopelessly overpowered by a combination of human engineering, machine learning, and massive-scale computation. When the media documents his loss for public consumption, his wife tells him “don’t look at the Internet,” lest he see criticism from his peers regarding his play. But Hui quickly recovers, in his own colorful and good-natured way. He eagerly becomes a collaborator, working with DeepMind to further evaluate and strengthen AlphaGo, in preparation for even stronger opponents. “[I] do my best to protect human intelligence,” Hui says with a smile, after helping to accomplish the opposite.
The film crescendos to a five-game matchup between AlphaGo and Lee Sedol, one of the world’s best Go players, in Seoul, South Korea, in March 2016. At the time, the match was widely covered in the media, and the outcome won’t surprise any theatergoers. Nonetheless, the film succeeds at building suspense. The AlphaGo team is anxiously making final tweaks, trying to diagnose some rare but potentially embarrassing flaws. As a world champion, Sedol struggles with unusual pressure. After all, he is not just representing himself; he is representing humanity.
The film succeeds at overcoming another obstacle: making Go feel like a spectator sport even to the uninitiated. The matches are slow, and unless you are a Go expert, the action is difficult to discern. But with careful editing, the film conveys AlphaGo’s actions through Sedol’s reactions. Hours of Sedol’s penetrating thought are distilled into a series of nervous twitches, gasps, and frustrated expressions that reveal a man, at the pinnacle of his craft, facing a new kind of challenge.
In the AlphaGo control room, we see a different picture. The room is brimming with emotion, but from the creators rather than the creation. There are moments of anxiety and joy as the DeepMind scientists monitor various diagnostics, trying to analyze what AlphaGo is thinking. Oblivious to it all, AlphaGo presses on, through good moves and bad.
There are some added theatrics. Earlier, as the DeepMind team prepares for the flight to Seoul, AlphaGo (version 18) is loaded onto a laptop and tucked into a briefcase—as if Sedol could just open the laptop, boot up AlphaGo, and begin the match. In reality, AlphaGo requires a veritable computing arsenal, running on 1,202 CPUs and 176 GPUs for the match against Hui, and possibly more for the match against Sedol.
In South Korea, the public spectacle draws comparisons to the Super Bowl, a comfortable subject for Kohs’s directing, given his time at NFL Films. At the airport, the DeepMind scientists arrive like a championship sports team, swarmed by flashing cameras. In the streets of Seoul, the games are streamed live on giant billboards. There is play-by-play commentary. There are crowds of fans.
After the match is clinched, there is sadness as a new reality sinks in. “I want to apologize,” Sedol says, with palpable disappointment. For the DeepMind team, victory is bittersweet. “I can’t celebrate,” says Hassabis.
Sedol, Hui, and the others ultimately find joy, and an appreciation for the new and beautiful. Go has been studied for thousands of years. Now there is a chance to see it anew, through AlphaGo’s eyes. It will “show humans things we never discovered,” Sedol says.
For the rest of us, AlphaGo's triumph is less personal. Most other human pursuits (literature, music, comedy, cooking, etc.) are still beyond the reach of current systems, and it may be a long wait before AI brings new meaning to these domains. The technology behind AlphaGo works best on specific tasks with very large amounts of training data. AI will need new approaches to tackle more open-ended pursuits from small amounts of experience, as people are able to do.
Nevertheless, when AI took the crown in chess and Jeopardy! it didn't remove the magic for millions of players and fans. With Hui and Sedol as our guides, AlphaGo encourages us to embrace the changes still to come.
Brenden Lake is a Moore-Sloan Data Science Fellow at New York University.
The inside story of how ChatGPT was built from the people who made it
Exclusive conversations that take us behind the scenes of a cultural phenomenon.
ChatGPT is about to revolutionize the economy. We need to decide what that looks like.
New large language models will transform many jobs. Whether they will lead to widespread prosperity or not is up to us.
Sam Altman invested $180 million into a company trying to delay death
Can anti-aging breakthroughs add 10 healthy years to the human life span? The CEO of OpenAI is paying to find out.
GPT-4 is bigger and better than ChatGPT—but OpenAI won’t say why
We got a first look at the much-anticipated big new language model from OpenAI. But this time how it works is even more deeply under wraps.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.