3

This question was inspired by watching AlphaStar play Starcraft 2, but I'm also interested in the concept in general.

How does the AI decide what build order to start with? In Starcraft, and many other games, the player must decide what strategy or class of strategies to follow as soon as the game begins. To use a Starcraft-specific example, one must decide to 6-pool Zerg Rush before any scouting information has been gathered. Delaying the rush to wait for info means the opponent will be stronger when the rush arrives; the opponent may even discover the rush and prepare a dedicated counter.

This is not limited to deciding between a risky early all-or-nothing attack. Some long-term strategies also preclude others. Terran players must decide early on how heavily they will invest in mech units. They can focus on biological units like marines, or vehicular units like siege tanks and hellions. Going equally into both, however, often means a weaker army overall, because you must spend resources on the overhead costs of both tech trees. You must upgrade your vehicle weapons as well as your infantry weapons for instance, meaning less resources can be spent on more units. Suffice to say, Terran players usually must decide very early on what they will focus on.

How can AI make these kinds of choices given incomplete and often uncertain information?

Ryan_L
  • 131
  • 1
  • General approach is "information sets" - factorizing state space by observable: https://en.wikipedia.org/wiki/Information_set_(game_theory) – mirror2image Jan 01 '20 at 06:37

0 Answers0