Artificial intelligence is not just about teaching computers how to play games. Research has already extended into many areas of human existence.
Artificial intelligence is an all-purpose term to describe non-human learning or problem solving.
Its development came with advances in computing technology. Increases in speed and functionality helped computers edge closer to mimicking human cognition.
Poker players often hear about AI in the context of game-playing programs.
Researchers have for years used games as testing grounds for a program’s “intelligence.” First for simple mathematical calculations, then for more complex types of reasoning.
It started with checkers
In the mid-1950s, researchers at several top US universities created computer programs smart enough to play checkers.
Early programs were functional. In in some cases they even provided genuine competition for human opponents. But it wasn’t until 1994 that a checkers program called Chinook defeated champion player Marion Tinsley.
And it wasn’t until 2007 that researchers claimed definitively that checkers had been “solved.” Albeit in the sense that the top players could only ever reach a draw versus Chinook.
Chess, Go… and Jeopardy?
Meanwhile in 1997, IBM developed a chess-playing program called Deep Blue. Its mission was to defeat champion chess player Garry Kasparov.
More recently, AlphaGo, created by Google’s DeepMind Technologies, beat a professional human player. This time at the board game Go (in 2015). This was another significant breakthrough in AI.
IBM’s Watson software followed in 2010. This time the software was capable of winning a televised match of Jeopardy! versus former champions Ken Jennings and Brad Rutter. (One wonders how Watson would do against James Holzhauer.)
Work is being done to create AI-powered “players” of complicated, multi-player online games as well. One program called OpenAI Five defeated champion Dota 2 players just a couple of months ago.
Like checkers, both chess and Go are so-called “perfect information” games. That means competitors have access to all the available information.
Why AI poker is different
That’s not the case with poker. Instead, poker is an “imperfect information” game. Some information is concealed from view, such as hole cards in Texas hold’em. It makes poker a goal for AI researchers hoping to create a program capable of competing with or even beating a human player.
Back in the early 1960s, computer scientist Nicolas Findler began developing poker-playing programs, and published his findings in journals.
The idea caught on in film too.
The 1972 science fiction movie Silent Running features a similar scene. A crewman aboard a spaceship programs two “drones” or robots to play poker with him.
After he teaches them the rules of the game they play a couple of hands. When one of the robots wins a pot, the crewman laughs excitedly, the outcome of the game appearing to signal the robot has indeed managed to display human intelligence.
Humans and machines play heads-up
In the film the game is five-card draw. Later in the 1980s Mike Caro used an Apple II computer to create a program that could play fixed-limit Texas hold’em, then also no-limit hold’em.
Calling his program ORAC (his name spelled backward), it earned widespread attention when matches were staged at the WSOP. It was there the program took on Main Event champions Tom McEvoy and Doyle Brunson.
Another match between ORAC and Bob Stupak was aired on the TV program Ripley’s Believe It or Not.
Researchers at several universities continued the work.
In 2007, the annual meeting of the Association for the Advancement of Artificial Intelligence in Vancouver, Canada, decided to stage a contest. It was called the “First Man-Machine Poker Championship.”
It pitted poker pros Phil Laak and Ali Eslami against a program called Polaris, developed by researchers at the University of Alberta — the same group that had developed Chinook.
The game was fixed-limit hold’em and the human players won. But a year later came another match involving Polaris 2.0 and a different set of human opponents. This time the computer won.
The Alberta researchers continued their work, eventually creating a new poker-playing program called Cepheus. It was proficient enough at heads-up limit hold’em for the research team to describe that game as “solved” (per a qualified definition of the term).
Meanwhile another group working at Carnegie Mellon University created a program called Libratus. It played heads-up no-limit hold’em, and in early 2017 defeated four professional poker players in a NLHE match.
AI goes beyond playing games
Tuomas Sandholm, a professor of computer science and one of the CMU researchers, has explained how the point of such study is not simply to create programs that can win at poker. Instead, it is to develop an AI capable of reasoning in situations where the available information is partial. That is to say, an AI that can think and respond like humans do in most situations we face.
Sandholm explains how Libratus could also be employed: “in any situation where information is incomplete, including business negotiation, military strategy, cyber security and medical treatment.”
In other words, the advancement of AI will continue to be marked by programs achieving new, remarkable milestones in various games — including poker. But much greater significance will be how such advancements will affect other areas of our lives.
Indeed, decades from now AI-related research will likely have at least some impact on just about every aspect of life that a person can think of. Or that a program will “think” of.
Image via www.wpnsrus.com.
