Google’s mission to excel in the field of Artificial Intelligence is once again bearing its fruits with the launch of AlphaGo Zero. According to a report by The Telegraph, AlphaGo Zero has set a magnificent record of grasping thousands of years of human knowledge in just 40 days, considered as a major breakthrough, ever in the field of artificial intelligence. AlphaGo, the previous version of AlphaGo Zero surprised the world last year by beating world champion Lee Sedol (legendary South Korean champion) in the “Go” game. Go is an ancient Chinese board game which has complex rules and requires high intuition and strategy. While AlphaGo was trained and programmed to master the brilliant moves, the new AlphaGo Zero does all this by itself, learning from scratch. AlphaGo Zero is only taught how to play the game, with no additional instructions given. Over the time, it learned the best moves by playing the game against itself, multiple times.
Beyond Playing Games
Today, Artificial Intelligence is making an incredible progress in a wide variety of domains. Critics often label most AIs as “narrow”, as these focus only on single tasks like translating languages, recognizing faces, speech or voice. Can AI outperform humans at different and difficult tasks? This is the question that haunts Demis Hassabis, the CEO of DeepMind, who heads the creators of AlphaGo Zero. Hassabis believes that AlphaGo’s descendants will work alongside humans as scientific and medical experts. “It is not just about winning the game of Go, AlphaGo Zero is also a big step for us towards building these general-purpose algorithms”, remarks Hassabis.
How it works?
The program gathers and develops its skills through a process termed as reinforcement learning. A group of software “neurons” are connected together to form an artificial neural network. This is the core of the program. During each turn of the “Go” game, the network observes the position of the pieces on the board and calculates its next move towards winning the game. After each game, the neural network is updated with the learnings gained. This enables the AI to evolve as a strong player over a period of time.
AlphaGo Zero research team hopes that these kinds of algorithms and their future versions will constantly work with advancing the boundaries of science and medicine, and will soon start working in tandem with smart humans.