For hundreds of years, if not more, man has been fascinated with the idea of making an automated device able to interact with us like another human. Playing a game such as chess, checkers, or Go is one way to prove a machine’s thinking ability, or at least make it appear as such.
Here are five AI gaming machines that have either beaten us or are well on their way.
IBM Deep Blue Chess Supercomputer
In perhaps the first great blow to humankind’s idea that its intelligence would never be matched by a machine, IBM’s Deep Blue supercomputer beat chess world champion Garry Kasparov in 1997. This event had been in the works for many years, however. Claude Shannon wrote an important paper on the structure of this type of program in 1950, and “good” amateurs started to be beaten in the 1970s.
Though extremely impressive, it should be noted that chess has not yet been entirely “solved,” a situation where no matter what it is theoretically impossible for a human to beat the machine. Simpler games like tic-tac-toe, connect four, and checkers have been solved, making the chance of a human victory perhaps out of the range of possibility.
IBM Watson – Jeopardy
As challenging as chess can be to automate, it has structured rules. Each piece moves in a set pattern, and the objective of the game doesn’t change. The game show Jeopardy, however, presented a different challenge. Instead of rules and responses to moves, there are questions pulled from literally anywhere. A machine would have to both translate what was required into machine language, then search a massive database of information to determine the correct answer.
IBM Watson in 2011, containing the equivalent of 200 million pages of locally-stored information, defeated champions Brad Rutter and Ken Jennings. These two were arguably the all-time best Jeopardy contestants, but Watson defeated both handily.
Today, this “gaming AI” has been transformed into a system to analyze information in its natural state, and serve it back to humans as answers to real-world questions.
Although largely unknown to Western observers, Go has been played in East Asia since at least the fourth century B.C. Though the rules are simple and all the pieces are the same, Go has many more possible scenarios than chess. There are 20 possible opening moves in chess, whereas in Go there are a staggering 361. As the game progresses, the number of situations expand geometrically, create a game that is easy to learn yet incredibly difficult to master.
As with chess, some thought Go couldn’t be mastered by a computer. Google’s AlphaGo program in March 2016, however, beat Korean Lee Sedol in 4 out of 5 matches. The machine was trained – perhaps “coached” is a more appropriate term – by humans to respond to different types of opening moves, then play against itself to figure out an appropriate strategy.
Claudico Poker AI
Poker presents several unique problems for AI, and at the time of this article, humans seem to still have the edge. Carnegie Melon developed a poker AI called “Claudico” that lost to human competitors by a wide margin in a 2015 competition. Like the previously listed AIs, it was competing against extremely good competitors, but wasn’t yet able to measure up.
According to Noam Brown, who worked on the project, the number of possible game situations exceeds the number of atoms in the universe. Nevertheless, this AI has been trained in a similar manner to the new Go champion. It was programmed with the rules, then spent months playing hand after hand against itself, seeing what strategy would work best.
Soccer is much different from the aforementioned games, but the concept is simple – kick a ball into a goal – yet it is difficult to master. Soccer seems like it is an ideal sport for robots to play, and the RoboCup competition has been established with the idea that a team of robots could be able to defeat human Word Cup champions by 2050.
Though this seems like a lofty goal for a short 34 years from now, consider that 34 years ago IBM introduced the 286 processor, running at 4MHz to a top speed of 12MHz. Modern smartphones can top that by nearly 1,000 times.
This year’s Cup takes place from June 30th to July 3rd, with a symposium on July 4th. In Leipzig, Germany. Check out a few highlights from last year’s competition below.