AlphaStar

AlphaStar is an AI system released by Google DeepMind in 2019, built to play the real-time strategy game StarCraft II at a professional human level. Unlike games with fixed rules such as Go, StarCraft II requires reading an opponent's moves in real time while simultaneously managing resources, placing buildings, and controlling units — which made it a far more complex challenge by the standards of the time.

AlphaStar wasn't trained by having humans spell out correct answers one by one. Instead, the AI played an enormous number of matches against itself, gradually favoring the actions that led to wins more often. This trial-and-error approach to maximizing reward is called reinforcement learning, and AlphaStar became a landmark example showing that this method works even in a real, complex game environment.

The AlphaStar project has recently resurfaced in the news not because of the game itself, but because of what happened to the people who built it. The core team behind DeepMind's achievement moved together to another company, and the name AlphaStar is now being mentioned again — almost like a line on a resume explaining who was behind the original breakthrough.

Articles tend to mention it in phrases like "people who worked together on DeepMind's 2019 AlphaStar project," using it to introduce the people involved rather than the AI itself. It's easy to mistake AlphaStar for a new product or service, but here it refers to a research project that was already completed and released back in 2019 — the actual news is that people from that project have moved to a new startup.