Man vs. machine – Yesterday, on the Ides of March, the fifth game of the Google DeepMind Go challenge ended with a resignation from Lee Sedol, the Korean Go (or baduk, as it is called here) champion who stood in for humanity in this battle of man versus machine. Lee Sedol himself didn’t see it this way, though; when the series was decided after his third straight loss to AlphaGo, his machine opponent, he said that AlphaGo had not beaten humanity, it had just beaten Lee Sedol.
The series may have been a personal challenge for Lee, but I think a lot of people were happy to see him win the fourth game after a brilliant move that seemingly threw AlphaGo for a loop. After the third game, the Korean news was reporting that Lee Sedol had finally found a weakness in AlphaGo’s game, and when he managed to exploit that weakness in the the fourth game, it did indeed seem like a victory for humanity. I was curious to see if he could pull out a win in the fifth game as well. If he had been able to do so, I think a lot of people would have considered the series something of a victory for Lee Sedol, even though it was technically a loss. He would have gotten AlphaGo’s number, so to speak, and had the upper hand. That didn’t happen, though, and computers can add another human game to their trophy shelf.
I know next to nothing about Go. I know the basic concept of the game—which is actually pretty simple—but even the most rudimentary aspects of strategy escape me. Yet I was captivated by this series, and I even sat through some of the matches on television, despite having next to no idea what was going on. I did on occasion tune into the livestream of the fifth game on YouTube yesterday, and I think I learned a lot more by watching that than I did by what I saw on Korean television. The Korean commentators assumed that viewers knew what they were talking about, so my wife and I sat there completely lost for most of the time. It was like listening to a foreign language. The commentary on YouTube, though, was done by an American Go player, and he focused a lot more on explaining things. He still used a lot of jargon (mostly Japanese terms for formations and moves), but I felt like I was able to understand things slightly better.
I’m not really as interested in Go itself, though, as I am in the challenge of man versus machine. I think I would have liked to see Lee win, but I’m not too upset about his loss. I certainly don’t think that this means computers are threatening to conquer us or anything like that. It is, after all, just a game. More specifically, it is a puzzle. Strategy games with perfect information (meaning that nothing is ever hidden from either player) and no element of chance—games like chess, checkers, Go, and even tic-tac-toe—are essentially puzzles that can theoretically be solved. In this case, “solved” means that there is an optimal strategy by which one player can always win, or a strategy by which either player can force a draw. It also means that, assuming perfect play, the outcome of the game can be determined at any point along the way. Tic-tac-toe, of course, has already been solved—the number of possible moves is so limited that even children can figure out the optimal strategy. Even a more complicated game like checkers, though, has been solved as well. Chess is even more complicated, and Go is far more complicated than chess. Theoretically, though, with enough calculating power, even Go could be solved.
This is why a machine victory over humans in pure strategy games doesn’t really bother me. Pure strategy games, boiled down to their essence, require skills that computers happen to be very good at, things like calculation of permutations and memorization and utilization of patterns. As a result, AlphaGo plays at a much higher level than any of its creators; at one point during the last match, one of the AlphaGo team joined the commentators, and he said that none of that were really that great at Go. This came up during a discussion of the fourth match, which AlphaGo lost. When the program blundered after Lee’s brilliant move, the AlphaGo team didn’t know if it was glitching or playing properly. That was something they could only try to figure out in post-game analysis.
What might sound a little scary is the fact that AlphaGo learns as it plays, becoming better with each passing match. It is, in other words, an artificial intelligence. And, as we know from films like Terminator and The Matrix, artificial intelligence is bad news for humanity. The idea that computers can learn and improve themselves without human intervention is terrifying. And I think we all know how badly we treat our machines—just imagine what they would do if they started to think for themselves! While science fiction can be fun, though, it is still just that: fiction. One thing that is important to note in both of these films is that the AI is not just intelligent, it is sentient. That is, what started as an artificial intelligence evolved into artificial consciousness. The two are not even remotely similar, and there is no logical reason for one to develop into the other. That would be like a horse spontaneously gaining the ability to talk (with the notable exception of Mr. Ed). In the case of AlphaGo, of course, we have even less to fear, since it is a program designed for a very specific and limited purpose. It’s not like it is going to learn how to cook the perfect egg or file my taxes (unfortunately).
Of course, I don’t think anyone—at least not anyone intelligent—fears that AlphaGo’s victory over Lee Sedol heralds a dark new age for humanity. I just thought it was an interesting opportunity to write a journal entry and ruminate on a few things that have been bouncing around in my head. I think it’s natural to be a little disappointed that Lee was not able to win—humans are clannish creatures, and we root for our own. But although I am disappointed, I also find myself amazed that Lee was able to sit at that Go board and play five solid games. If you think about how much computing power AlphaGo had access to (the program uses well over a thousand CPUs in a distributed network), it is pretty impressive what Lee was able to accomplish with his single brain. And his victory in the fourth game showed that the computers haven’t beaten us entirely just yet. So I, for one, am not quite ready to welcome our Go-playing computer overlords.