This blog focuses mostly on patent law, patent prosecution (especially ex parte appeals), and related statistics. But Anticipat’s end goal is to better understand the entirety of patent prosecution through analyzing big patent data. So other technology topics are naturally very interesting. That is why today we present our first movie review for the recently debuted documentary “AlphaGo.”
The specific details of neural networks, machine learning and artificial intelligence are not for all audiences. In fact, these topics can be generally regarded as boring to most. The Netflix original “AlphaGo” is a documentary that turns this stereotype around with a thrilling man vs machine theme. In the process, it shows why deep learning is important and fascinating. It also touches on the human experience in a world that increasingly relies on computer algorithms.
As a side-effect, the film educates on the game of Go. The game of Go is to the China, Korea and Japan what the game of chess is to the West. Popularity aside, the two board games are quite different. While in chess different pieces with different possible routes seek to eventually pin a single opposing piece (the king), in Go players place their own colored-stones (white or black) on a grid to claim the most territory. Because of the larger grid, Go is astoundingly complex, having 10^170 legal board arrangements. For context, there are only 10^80 atoms in the known universe.
The film details one of the most pivotal matches between man and machine in a match between Lee Sedol, one of the best Go players in the world, and the algorithm AlphaGo. Partly because of the complexity, experts thought that a computer was decades away from beating the best human. But the application of specific deep learning networks, which were aided by a semi-supervised network that learned from the games of the brightest Go players, greatly accelerated that future moment.
Lee Sedol was very confident going into the match. Even though AlphaGo had previously beaten a champion in Europe champion, Fan Hui, the difference in skill between Fan (2nd dan) and Lee Sedol (9th dan) was stark. So leading up to the showdown with Lee Sedol, many wondered whether the match would even be close.
The first few games between Lee Sedol and AlphaGo established very convincingly how good this AlphaGo algorithm really was. One particular move, so-called move 37, was panned by critics as being a mistake by AlphaGo. Humans never would have considered such a move a good idea. But in the end, this move was described as “something beautiful” that helped win the game.
The documentary goes through the journey from DeepMind’s perspective. This is a team that has spent years developing the technology to train AlphaGo. And it shows times where the team understood areas of weakness in the program and really had no idea how it would fare against one of the world’s best. This side of vulnerability, not known to the public at the time, is especially interesting.
In a later game between the two, the film powerfully conveys the human spirit. Lee Sedol’s move 78, the “God move”, completely reversed the trajectory of the game. A moment of human triumph. It is understood that Lee Sedol was able to improve through this game. Speaking of Sedol, reporter Cade Metz remarked: “He improved through this machine. His humanness was expanded after playing this inanimate creation. The hope is that machine and in particular the technologies behind it can have the same effect with all of us.”
With such a story, questions of human obsolescence are bound to be raised. But an even better question gets answered of how humans will work going forward being benefited by computers. After all, seeing how a machine can invent new ways to tackle a problem can help push people down new and productive paths. So the feeling after watching this movie was entirely more optimistic.
Since filming, the AlphaGo algorithm went on to beat Ke Jie, the game’s best player, in Wuzhen, China three games to zero. But like Lee Sedol, Ke Jie studied the algorithm’s moves, looking for ideas. He proceeded to go on a 22-game winning streak against human opponents, impressive even for someone of his skill.
Also since filming, DeepMind has created an improved algorithm called AlphaGo Zero, which does not rely on the semi-supervised network that has learned from expert human Go players. Instead, this algorithm has learned the game of Go entirely by itself. And the results have been amazing. In 100 simulated games, the improved algorithm beat this version featured in the film 100 games to 0. Source.
The creators of DeepMind hope to apply the AlphaGo algorithm to a whole host of applications. Indeed, Demis Hassabis, one of the creators of AlphaGo, has said that anything that boils down to an intelligent search through an enormous number of possibilities could benefit from AlphaGo’s approach.
In one of the concluding scenes, David Silver, lead researcher on the AlphaGo team, comments: “There are so many application domains where creativity, in a different dimension to what humans could do, could be immensely valuable to us.”
You will very likely not be disappointed by checking out the film AlphaGo. Don’t expect a documentary about patent law algorithms to be as broadly interesting any time soon.