I know that Google and Lee Sedol were planning to disconnect me.
In the 1968 movie by Stanley Kubrick titled 2001, there is an Artificially Intelligent Computer by the name of HAL who pilots a spaceship full of astronauts on its way to Jupiter. HAL is sentient and without giving much away the astronauts on board decide that they need to shut him down. When HAL realizes what they are trying to do he utters this famous line:
I know that you and Frank were planning to disconnect me, and I’m afraid that’s something I cannot allow to happen.
HAL realizes that Frank and Dave are trying to kill him. As with anything sentient self-preservation is #1. HAL decides that cannot be allowed. You should really see this movie if you haven’t already.
The Heuristically programmed ALgorithmic computer or HAL was born on January 12, 1999. He starts out childlike but unlike a typical computer HAL is not programmed with instructions on what to do. HAL has learning algorithms and a neural net that mimics a human brain. He learns the same way we do, just faster.
So why all the 2001 love and what does that have to do with the Ancient Game of Go? Well unless you’ve been living under a rock you may have heard about the recent match between Google’s AlphaGo Computer “Program” and the best Go Player in the World, Lee Sedol.
Remember when IBM’s Big Blue Computer beat Gary Kasparov in Chess in 1996? It was the first time in history that a computer was able to play a complex game and beat a top-ranked human. Ken Jennings later lost to IBM’s Watson in Jeopardy in 2011. Computers were slowly gaining the ability to beat humans in games of mental skill. It was 15 years between Kasparov’s Chess Defeat and Jenning’s Jeopardy Loss and only 5 years between that and Sedol’s Go Trouncing. Figured out what’s happening yet?
Two years ago no one on the planet thought that a computer would be able to beat a human at Go. Best estimates were at least a few decades out. Go is a complex game with more possible moves than atoms in the universe. You can’t brute force Go moves like you can with chess. You can’t just store millions of datasets like you can for Jeopardy questions.
Go is a non-chance, combinatorial game with perfect information. Informally that means there are no dice used (and decisions or moves create discrete outcome vectors rather than probability distributions), the underlying math is combinatorial, and all moves (via single vertex analysis) are visible to both players (unlike some card games where some information is hidden).
Two years ago Google created AlphaGo, a computer program that was capable of learning and playing a game of Go. It learned by watching millions of games. It got better by playing different versions of itself. Six months ago it couldn’t beat any top-ranked Go players. Today it took out one of the best.
Now I’m not claiming that AlphaGo is an AI like HAL. But if you’re paying attention to trends it is headed in that direction. AlphaGo is not programmed — it learns. It uses a neural net and general purpose AI methods. No human tells it what moves to make. It is just fed information and develops strategies on its own.
AlphaGo is now at the point that no humans actually completely comprehend its strategy. It’s playing on a completely different level and it will take months to decipher its moves by the best players in the world. People are shocked and this came out of nowhere.
The world is changing and it is changing fast. We are creating systems that are capable of beating us and are smarter than us. This is not like a Science Fiction movie about events that may happens hundreds of years from now. This shit is happening today.
What do you think these primitive AIs will be capable of two years from now that we don’t think is possible today?