AI in reinforcement workouts is usually set to maximize your chances of winning, but this is not necessarily a strong workout. Software can only be distinguished under a narrow set of conditions and left open for operation. Like human experts, DeepMind's new approach has one of the AI's focus on exploiting the other's weaknesses. AlphaStar has gradually learned to try a greater variety of strategies that could do more to counteract unconventional, heavily exploited tactics (also known as cheese)
Technology still has its limits. For starters, it takes a lot more training than a person to meet a comparable skill level. This is no small feat, given the complexity of StarCraft's and it's good for DeepMind's long-term plans. As with previous studies of the game company, the ultimate goal is to translate AlphaStar's progress into real-world applications. Better-trained AI can help self-driving cars and robots deal with unusual situations that they would not otherwise be prepared to handle.