The AlphaGo AI program has become completely self-learning

Over the course of this year and last year, the AlphaGo program has repeatedly beaten professional Chinese Go players, including world champion Lee Sedol. This result was preceded by a long study of matches played by members of the worldwide Go game association. After that, the program played with itself, simultaneously analyzing all the moves. It was this algorithm that contributed to AlphaGo’s absolute victory in competitions with professionals. AlphaGo managed to surprise even experts: they expected to record a similar result no earlier than 2020. Recently, Google's division, which develops artificial intelligence systems, introduced the next version of AlphaGo - AlphaGo Zero. The updated application beats its predecessor “dry”, with a score of 100:0. The new version of the program no longer needs a human “teacher”; the learning process takes place independently, based on its own steps and their analysis. Before the moment of absolute championship, AlphaGo Zero completed 29 million games in just three days; for many recognized masters, this took decades. The previous version of AlphaGo was equipped with two independent neural network learning modules. One module was responsible for the current situation on the game board, the second was thinking through the next move. The new program has a single neural network that performs the above actions simultaneously, so the results of its work are more effective. Plus, AlphaGo Zero no longer limits itself solely to human knowledge, but goes beyond it, thinking through its own unique steps. In the history of the development of intelligence, this fact is decisive.