The notion that advanced computer programs are smarter than us is neither new nor surprising. There are, of course, plenty of areas where humans are irreplaceable, at least for the time being, but generally, artificial intelligence (AI) surpasses even the most exceptional human brain in many ways. This evidently includes playing games.
While it is surely an entertaining thing to see machines competing against people, the coders and computer scientists do not do their job entirely for fun no matter how much joy it brings them. There are usually some implications involved with working on a subject, and the history of AI versus men with respect to gaming has an interesting and meaningful story.
In 2011, IBM’s Watson got the better of the best Jeopardy! contestants, winning over $77,000 dollars and leaving its human rivals Brad Rutter and Ken Jennings far behind. This well-known and documented clash was one of the early signs that helped everybody to realize what machines could be capable of in the future. Since then, AI has successfully engaged with us on a few occasions in which it prevailed over humans including beating master Garry Kasparov at chess, some of the best eSports players at Dota 2, the Go champion and now, poker’s best in early 2017.
AI’s achievements in Go have been made possible by Google’s Alpha Go. However, its latest version Alpha Go Zero went at least one step ahead in the matter. Instead of simply mastering the presented set of rules, it managed to teach itself the game, starting totally from scratch with no human background (features, examples, data). That kind of learning is being called “tabula rasa” (blank slate) learning, which is clearly a more effective way to absorb and make the use of the knowledge than just using given information, especially if some unexpected errors or faults in data occur. This could have implications for many industries such as travel, logistics, medicine and manufacturing.
An even bigger deal for scientists is their success with the computer system called Libratus, which won an experimental battle with some of the best poker players. Anybody familiar with the rules and variations of the game of poker, as well as its general flow and detailed intricacies, is surely aware of the fact that it is not exactly trivial entertainment. The play is a result of numerous factors involved in many possible scenarios where your skills are put to the test in different situations such as how will you handle your emotions or deal with unexpected turns of events while managing incomplete information. This is why the significance of AI being good at poker really makes waves.
As opposed to chess or go, poker is considered imperfect information games, which simply means that not all the information is out there (like your opponent’s cards, for example). Mastering it, however, is much harder. Not only was Libratus able to teach itself to play better poker thanks to interactions with humans, but it also managed to pick up on their subtle flaws, gain on them and ultimately win. This is like a computer version of bluffing and could potentially assist us in various areas of life whenever the facts are incomplete and not all knowledge is available.
We may be able to find a new applicability for the machines in potentially everything, even something that may not seem ground-breaking at first like enjoyable, friendly games. As we’ve seen more than once in history, remarkable inventions and big ideas were developed with much less, and gaming seems to be a great field to watch how computers think. It is our responsibility to make the most of this experience and above all learn. You can be sure that they will.