On Wired.com today was an exceptional article about the rivalry between Facebook researchers and Google researchers to develop the first Artificial Intelligence ‘Go’ Grandmaster. For those of you unfamiliar with it, ‘Go’ is the Eastern equivalent of Chess in the West. The difference between them is quite vast, however. The problem AI scientists face is the complexity of Go to Chess. According to the author, a player has on average 35 moves one can make, followed by 35 moves, etc, until the game nears its end. In Go, players have on average 250 moves possible per turn, followed by 250 moves, etc. This creates a unique issue: Checkers and Chess AI have all been based on decision-making algorithms, so they decide in advance what is the best possible move to make. The sheer exponential amount of moves to analyze renders this AI form incapable of competing effectively at Go, as the AI cannot calculate possible outcomes timely enough.
Enter ‘Deep Learning.’ Deep Learning is an AI form that approximates functionality of human neural networks, adapts to stimuli, and effectively ‘learns’ by visual patterns and spacial awareness of the game’s board. Essentially, it emulate’s natural thought processes and human brain activity while playing and being consciously aware of the board. Apparently as one plays a game, eventually the decision-making process lends in to experience, or visualizing the game-state and acting upon the context of the board to result in the best potential move. This form of AI varies greatly from what is known as “Tree Search,” exhibited by the Checkers and Chess AIs. How this works specifically is much like a process of elimination based on analysis of the outcome of one’s moves.
To this new form of AI, Go players often remarked that it felt oddly ‘human.’ The way the game is played by the AI, it makes natural moves based on context as opposed to algorithmic tree searching. The ultimate beauty of this breakthrough is its ability to work within the frame of other AI algorithms. In other words, deep thinking AI and Tree Searching can be coupled to create a smarter, or intuitive, AI. But this is still being implemented by the Google and Facebook teams. I’m curious to see the outcome of their friendly little tech competition.
As a table top gamer myself, I’m interested to see the implementation of such AI platforms in new and existing digital games. I honestly couldn’t tell you how exciting it would be to play a card game with an adaptive AI, and Deep Learning may hold the key. Moreover, I’m eager to see how this type of technology will be implemented into existing platforms such as Web Design and SEO. If an AI can simply look at a board and make a smarter play, surely it isn’t outside the realm of possibility that an AI could scour a few web pages and design something wonderful. At this stage I’m a little wary of the content-production value of an AI, but I certainly see the potential for it’s use in design and development. Lastly, SEO would be incredibly simple for this type of AI. I see a future where Cortana, Siri, or Alexa will build your web pages, and then fight internally among each other battling for rankings and digital presence.