The game in question was Go, a strategy game that is over 2,500 years old and believed to be one of the oldest continuously played board games. Played on a 19x19 board, this game has even more possible configurations than the number of atoms in the observable universe... or even chess!
With the constant articles about the inevitable AI takeover, it gave me an odd sense of satisfaction to see the supposed underdog win for once. This satisfaction quickly soured when I found out that a computer program was used to find weaknesses in the AI's strategy beforehand and help the player plan his counter strategy.
Apparently, the strategy used to beat the AI was at the level of an intermediate-player. Ultimately, an AI performs as well as its training, and this emphasises the importance of using an effective training set when training deep-learning systems. It also reminds us why so many companies offer their training sets as a service. On the other hand, it is equally important for there to be transparency in how these AI learn, so that more efficient training can be performed. With a better understanding in how to train AIs and the optimal training sets to use, fundamental flaws can be fixed rather than exploited.
Even though it might be a bit too early to place AI on a pedestal of perfection... I wouldn't bet on beating one in a game myself.
"due to the huge number of possibilities it is impossible for an AI to account for every scenario, leaving it vulnerable to unexpected moves"
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
We operate a free-to-view policy, asking only that you register in order to read all of our content. Please login or register to view the rest of this article.