AI related to gaming is not a new thing. We have seen AI beat humans in chess games, e-sports or even drone sports. Each of these AIs is developed to win to test its ability to beat humans. But SIMA is a different kind of video game AI. It is trained not to win itself, but to help human players win.
SIMA or Scalable, Instructable, Multiworld Agent is an AI developed by Google DeepMind. It is trained to complete any type of game such as open world or one that does not have a fixed storyline. SIMA can accept and obey user commands and understand the 3D world of video games. This AI is also not a feature or function built into the game because SIMA cannot access the video game code but is like an assistant to the main player.
To train this system Google partnered with eight developers such as Hello Games, Embracer, Tuxedo Labs, Coffee Stain and more. SIMA was also trained with videos of people playing the video games No Man's Sky, Teardown, Valheim, and Goat Simulator 3 and this AI played the games itself with the player's instructions. Google also recorded a video of two people playing, one controlling and one giving instructions for SIMA to understand how humans play.
According to Google DeepMind, SIMA now has 600 different skills such as turning, climbing stairs and opening the menu to access the map. SIMA is only capable of performing simple tasks but is believed to be able to directly complete a mission in any video game. In other words if you are the main player or the first player, SIMA will be the second player to help not to compete.