Could we program a cognitive computer to examine the problem of cancer, or the yearly flu outbreak, balancing the federal budget or just finding a way to reduce traffic congestion around the city?
John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology and government. He is currently the CEO of the Tech Writers Bureau, a group that creates technological thought leadership content for organizations of all sizes. Twitter: @LabGuys.
If you happen to have a free 30 hours or so, I would highly recommend watching Google’s AlphaGo program take on one of the best players in the world at the ancient Chinese board game Go. If you don’t have that much time, you could instead just watch the 6-hour third match, where the program wrapped up the best of five series. It’s literally history being made.
Some news outlets have covered this feat, but I don’t think many people understand how monumental this actually is. Back in 1997, when Garry Kasparov was beaten by IBM’s Deep Blue in chess, people were more excited about the future of computing.
But when you look at the underlying technology, the AlphaGo win is so much more impressive because it required a computer to actually think like a human to beat a human.
Deep Blue was a supercomputer able to use brute-force tactics to beat a grandmaster at the game of chess. Some say Deep Blue was programmed to specifically be able to beat Kasparov, and while I don’t know if that was true, it was likely within the realm of possibility for the program.
AlphaGo could not have been similarly programmed to take on Korean master Lee Se-dol, considered the best in the world. To beat a human at Go, a program has to really think like a human.
You see, chess and Go are fundamentally different games. In chess, each time a player moves, he or she has an average of about 20 options. With Go, the average is about 200. Because of the more limited options in chess, it makes it easier for a computer to store specific game variations and act accordingly to follow a winning strategy.
In chess, by the second move in the game, there are 72,084 possible game combinations. On the third move, it jumps to 9 million possible games. By the fourth move, it jumps again to 318 million possible games.
That seems like a lot, but a dedicated supercomputer could crunch those numbers within the time allotted for a turn in chess. Plus, Deep Blue could have had the most likely moves Kasparov would make based on his history set at the top of its memory stack, which leads to the accusations it was designed to beat one person.
You simply can’t do any of that with Go, which is played on a 19-by-19 square grid with two players alternatively placing white and black tiles. Players can capture their opponent’s tiles by surrounding them and turning them to their color. The winner is whoever owns the most tiles at the end of play. In terms of game variations, those who study Go say the possibilities are almost equal to the number of atoms in the universe.
To win at Go, and especially to beat the best human in the world, AlphaGo had to use superior pattern recognition and actual learned strategy, just like a human player. And AlphaGo played like a human, too.
If you listen to the commentators for match three, at times they seemed confused at the strategy the program was taking, which at different points was even risky, though it got away with it. They concluded the program must have been “showing off” for the third match to sweep the series.
The real victory here is the ability to show that thinking machines are starting to use their intelligence much better, and within proper context. AlphaGo used qualitative reasoning, which is one part of artificial intelligence. Wikipedia defines the new science as an area of research within artificial intelligence that automates reasoning about continuous aspects of the physical world, such as space, time and quantity, for the purpose of problem solving and planning using qualitative rather than quantitative information. This allows the intelligence to contemplate infinite possibilities, at least within structured rulesets.
Unlike true artificial intelligence, AlphaGo will never become fully cognizant. It’s doesn’t care if it wins or loses, and only “knows” reality within the terms of the game.
But think about how powerful that could be for humanity. Could we program a cognitive computer to examine the problem of cancer, or the yearly flu outbreak, balancing the federal budget or just finding a way to reduce traffic congestion around the city?
AlphaGo is hopefully the first of these superstar cognitive computers we will experience. It showed what was possible, and now we need to direct that technology at things which are more important than just a game.