Git Practice

Google’s MuZero chess AI reached superhuman performance without even knowing the rules

https://www.technologyreview.com/2021/09/29/1036329/podcast-how-games-teach-ai-to-learn-for-itself/

This article Discusses one of Googles newest technological developments in AI. Google has been pioneering this front ever since it aquired the company Deep Mind, based in london. This Company was the one that actually created the AI called AlphaGo that beat the best Go player in the world after many months learning to play on its own using only the rules and by playing against itself over and over again. After that, Google had bought the company making it a child company. Typically algorithms will decide the best method by calculating and comparing the outcome of all the methods, but this AI decides the best move to make based on what seems better, almost Human-like intuition. Now this time, an AI called MuZero was not even informed of the rules of chess, it was not able to make illegal moces but it coulf think about them. This is similar to how we as humans do and think, granted humans think about doing illegal things but we consider the repurcussion and results. This was even more fascinating than AlphaGo because it never was told the rules it began to learn itself based on its previous games. This AI suprisingly beat the original AlphaGo while using less computation as it began to think even more like a human. It also learned to play multiple games. This is a breakthrough in AI, as we are always learning how to make tasks simpler, so less computation and higher performance was a good sign of the capabilities of AI in the future and how we're getting closer to human-like AI. But still we are very very far from it ever being like a human.