AI has been successful in beating humans at various games such as checkers, chess, Go, poker, and even Jeopardy!. One of the most recent and impressive AI victories was against the world champion Go player Lee Sedol. Checkers and chess were some of the first games that AI was able to beat humans at. In these games, there are a finite number of possible moves which makes them easier for computers to calculate all the potential outcomes and find the best move. However, Go is much more complex than either checkers or chess with a nearly infinite number of potential moves. This made it much harder for AI to defeat human players – until recently.
In March 2016, an AI program called AlphaGo defeated Sedol in a five-game match by four wins to one. This was a significant achievement as Sedol is considered one of the best Go players in history. AlphaGo used a combination of machine learning techniques to learn how to play Go from scratch and then defeat Sedol – something that many experts thought would be impossible for AI to do.
Poker is another game where AI has been successful in beating humans. In 2015, an AI program called Claudico beat four professional poker players in Texas Hold’em – another game
Go: DeepMindS AlphaGO VS Worlds Top Five Players
Artificial intelligence has been making headlines for its successes in a number of domains in recent years. In March 2016, an artificial intelligence program called AlphaGo made history when it defeated the world’s top Go player, Lee Sedol, in a five-game match. This was a significant achievement because Go is considered to be a much more complex game than chess, and many experts had thought that it would be many years before an AI program could beat the best human players.
In the months since AlphaGo’s historic victory, DeepMind (the company behind AlphaGo) has continued to develop its algorithm, and in October 2017 they released a new version called AlphaGo Zero. This new version of the software is even more powerful than the original; not only can it beat all of the world’s best human players, but it can also defeat the previous version of AlphaGo (which itself was undefeated against all other Go programs).
To understand just how impressive these achievements are, it is helpful to understand something about Go and its complexity. Go is played on a board with 19 horizontal and 19 vertical lines (or “intersections”). Players take turns placing black or white stones on these intersections; the aim of the game is to surround more territory than your opponent. The game ends when both players have passed their turn, at which point any remaining empty intersections are considered to be surrounded by whichever player has more stones adjacent to them.
The rules of Go are very simple; however, because of the large number of possible moves (there are 378 points on a standard 19 x 19 board), and because there are no captures as in chess, Go is an incredibly complex game. In fact, according to one estimate there are more possible positions in Go than there are atoms in the observable universe!
This complexity makes perfect play by both sides impossible; even with perfect knowledge of all past games played by humans (and computers), it would still be impossible for either player to know what move to make next with complete certainty. Because of this inherent uncertainty, winning at Go requires not only tactical skill but also strategic planning and intuition. As such, it has long been seen as one domain where human intelligence might still have an advantage over machine learning algorithms.
Backgammon: BKG 8 vs Luigi Villa
In the early 1980s, Luigi Villa, one of the world’s best backgammon players, was challenged to a match by BKG 8, an artificial intelligence (AI) developed by a team of researchers led by Hans Berliner at Carnegie Mellon University. The match was played over several days in September 1981 and ended with BKG 8 winning 7-0.
This was not the first time that an AI had beaten a human player at backgammon (or any other game, for that matter), but it was certainly one of the most significant victories. For many people, it demonstrated once again that computers were becoming increasingly capable of outperforming humans at complex tasks.
Interestingly, Villa later went on to defeat BKG 8 in a rematch held in December 1981. However, this victory was short-lived as Berliner’s team soon developed an improved version of their AI which quickly regained the lead. In fact, over the next few years, AIs would continue to dominate human players at backgammon (and other games), culminating in Deep Blue’s famous victory over Garry Kasparov in 1997.
Poker: Libratus vs Four Top Players
In January of 2017, a computer program called Libratus became the first AI to beat top human poker players in a major tournament. The event was called the Brains vs. Artificial Intelligence challenge, and it took place at Rivers Casino in Pittsburgh. Libratus was developed by two Carnegie Mellon University computer science professors, Tuomas Sandholm and Noam Brown.
The tournament consisted of 120,000 hands of no-limit Texas Hold’em, with each hand played out over the course of 20 minutes. Libratus faced off against four human players: Jason Les, Dong Kim, Daniel McAulay, and Jimmy Chou. These four players are some of the best poker pros in the world; in fact, they have a combined total of over $19 million in live tournament winnings.
The humans started out strong, but Libratus quickly began to catch up. By the end of the tournament, Libratus had won $1.8 million in chips from its opponents; Les was the only human player who managed to finish with a positive balance (he won $80k).
So what made Libratus so successful? One key factor was its ability to identify patterns in its opponents’ play styles and exploited them accordingly. For example, Les tended to raise pre-flop when he had a strong hand; Libratus learned to identify this pattern and would often re-raise him when it knew it had a stronger hand itself. Similarly,…
“The best thing about playing games is that sometimes you get to beat humans.” -Unknown
The future is bright for games that beat humans. With the rapid pace of technology, it’s only a matter of time until games are able to beat humans at any and every game. Soon, we’ll all be playing against artificial intelligence and enjoying the challenge of trying to outsmart them.