In need of evolution: game theory and AI
This article reminds me a news: Google AlphaGo eventually won player Li Shi in the man-machine battle. This confrontation between human intelligence and artificial intelligence has set an unprecedented wave of attention to artificial intelligence around the world. AlphaGo is an artificial intelligence program developed by Google. The pattern of the winning and losing rate (that is, the value network), and finally through the Monte Carlo tree search to determine the optimal drop. At the same time, Google uses more than 1000 CPUs and GPUs for parallel learning and searching.
As we learned in class, game theory helps us do best decision. So how exactly AI connects to game theory? Shortly after the emergence of game theory, the field of artificial intelligence was followed by development. In fact, pioneers of artificial intelligence such as von Neumann and Simon have made outstanding contributions in both early days. Game theory and artificial intelligence are actually based on decision theory.
I think although it’s not that difficult to figure out when and why the paths of game theory and AI became convoluted, we overlook the restrictions AI, and in particular multi-agent reinforcement learning, has to face when following classical game theoretic approaches. Evolution will not happen automatically unless we try it, so it can offer its theoretical tools and practical advantages. Whether deep learning techniques can help solve large-scale game problems is also worth exploring. Perhaps we can’t prove that a deep learning-based strategy can form a certain equilibrium, but it may be close to the equilibrium strategy from the experimental simulation results.