Humans have always been defined by their creativity and intuition. Now, Deepmind’s artificial intelligence program, AlphaZero, owned by Google’s parent company, has just edged ever closer to grasping true human behaviour and tendencies, in what has been called a “historical turning point”. AI has, according to recent findings, generated its own way of chess playing.
Rather than using former experiences of winning or losing to statistically play its next move, AlphaZero bases its moves on a “nebulous sense that it is all going to work out in the long run”, according to experts at DeepMind. Their findings have been pubished in the journal Science.
AlphaZero lost just six of 1,000 games in competition with AI’s former chess master, Stockfish, winning 155 and drawing the rest. While Stockfish analyses roughly 60 million outcomes a second, AlphaZero searches a measly 60,000 outcomes a second. Like all chess machines except AlphaZero, Stockfish bases its moves on objective statistics rather than a feeling or intuition. AlphaZero creates its own tactics, generating its own style of play.
The AI began as a ‘tabula rasa’, or blank slate system, built with only the elementary rules of chess, learning victory through experience and intuition by playing millions of games against itself. Human brains learn in the same way, changing their behaviour based on former experience. By using a blank slate system, the team behind DeepMind are hoping that AlphaZero’s ‘brain’ could also be applied to tackle real world problems, developing medical treatment for diseases like Parkinson’s and Alzheimer’s.
Elsewhere in the world of AI, bots have been learned to dress themselves (awkwardly right now, but they’re getting somewhere), making £££ art and presenting the news.
