How AI and Machine Learning allows computers to play board games against us.
Earlier this month we featured a post that went into a little more depth explaining Artificial Intelligence, and how Machine Learning and Deep Learning work.
Following on from that, we thought we’d do a little “AI Month” and talk about two AI systems that went a long way to demonstrate the power of AI and, in a sense, functioned as watermark “proof of concepts” for AI technology.
Those two systems are IBM’s Deep Blue, which famously beat the chess grandmaster and World Chess Champion, Garry Kasparov, at his own game in 1997, and, more recently, Google DeepMind’s AlphaGo which beat a master Go player five-for-five at an ancient war strategy game—a game considered to be the most complex game in the world.
As a kind of follow-up, we thought you might want to learn a little more about these two systems and how they work.
First Day of School: Deep Blue
While, by today’s standards, IBM’s Deep Blue is nothing out of the ordinary for the time it was a benchmark for the potential of computers. Today one can play against computers equally as complex as Deep Blue from a mobile device a fraction of the size of the computer that played against Kasparov.
Deep Blue was artificially intelligent in the sense that it could do a task that a human could—in this case, play chess against a human—which is the most basic definition of AI.
However, the video highlights a very important distinction: Deep Blue did not learn how to play chess, but was rather told how to play chess; it didn’t necessarily ‘understand’ chess.
Deep Blue neither learned, nor did any “thinking” per se: its approach was to “brute force” the solution by computing every possible action to the given state of the chessboard at the time. Where humans can’t possibly ‘compute’ every single possible outcome, a computer can—thousands of them every second—and that’s exactly what Deep Blue did.
It’s the equivalent of a human never learning how each chess piece moves and then using that information to strategise accordingly against an opponent’s moves, but instead looking in a book that contains all the possible moves one can make given the different possible states of the board, every time a move is made, in order to find what the next best move would be.
In contrast, nearly 20 years later, the field of AI has grown exponentially to the point where Machine Learning and Deep Learning allow computers not only to teach themselves but can react and adapt to changing and unpredictable circumstances far more organically.
Growing up Fast: AlphaGo
What’s interesting about AlphaGo—and what sets it apart from Deep Blue—is how it learned to play Go. Deep Blue was programmed to do only one thing: play chess, and to do so it used raw computing power to analyze every possible move. AlphaGo, on the other hand, wasn’t programmed how to do anything: it learned how to play Go.
It did so by watching the game being played—interestingly enough, on the internet—and then playing it against itself hundreds of thousands of times and learning from its mistakes. This is a process that mimics closely how a human would learn how to play the game, or anything really, and is called ‘Reinforcement Learning’. The advantage of AI is that they can just watch and play way more things than a human can—there are only so many games a human can play in a single day, but not so for a machine.
AlphaGo’s programming was also streamlined over Deep Blue in another aspect: the way it played the game.
Go is one of the most complex games in the world and has thousands of billions of possible moves (10170 to be exact, which is more possible solutions than there are atoms in the Universe). AlphaGo, much like humans, uses experience to justify why the next move is logical, and the possible consequences of making that move.
AlphaGo does not know every possible move and permutation of a move like Deep Blue, simply because a feat that would require just too much processing to handle the necessary calculations. If it wins multiple times with a particular move when confronted by similar circumstances, it remembers and is more likely to use the move next time, much like a human would.
Graduation Day: Conclusions
Both Deep Blue and AlphaGo demonstrate the ability for computers to teach themselves how to do things themselves, ‘reason’ and learn from their mistakes.
Deep Blue and AlphaGo are just two examples; already there are many more upgrades and attempts—IBM’s Watson computer, for example, was trained to play (and won) the television trivia game show Jeopardy!
While AI playing games with humans might appear frivolous, it demonstrates the ability of AI programmes to show some kind of improvisation.
Since they are often put up against the most unpredictable of competitors (ie, humans) improvisation is a trait that is undeniably sophisticated and crucial for machines—a trait that might one day have wide usage in an unpredictable world.
More than just game-players, these demonstrations are game-changers; they can be used as the first concepts offering glimpses into the future utility of AI; these are in essence experiments in the core goals of AI research: General AI, or artificial intelligence that can do more than one task.
Digital Cabinet is more than just a paperless document management system. As we mentioned previously, our goal is workflow: we want to ensure that businesses keep up-to-date with technology trends and provide solutions that make running a business as simple and efficient as possible, with automated workflow as our number one priority.
To that end, we feel it is our duty to keep our ear to the ground and see how the latest and future technologies can be used and implemented to further improve and iterate on this vision. As we do that, we also want to keep our clients and readers informed too, so that you can start thinking about the potential of the future.
So, next time you feel like playing a game but there’s no one around, maybe consider turning to your computer—you might learn a thing or two, or maybe it will…
Bonus content: If you’re wondering about whatever happened to World Chess Grandmaster Garry Kasparov after his defeat by Deep Blue, well, the answer is that he never really forgot about AI…
You can find out more about Digital Cabinet at www.digitalcabinet.co.za