Google is training an AI to play video games — why this could be problematic

UPDATE: Mar. 15, 2024, 12:59 p.m. EDT This piece was updated to include additional information provided to Mashable by Google DeepMind.

Video games with computer-controlled characters like NPCs have been around for a while. However, Google’s AI research lab, DeepMind, has introduced something unique: SIMA, which stands for Scalable Instructable Multiworld Agent.

Unlike typical NPCs, SIMA isn’t programmed for a specific game. It acts as a generalist AI agent for 3D virtual environments, interacting with human players through verbal commands.

SIMA differs from NPCs

SIMA collaborates with gamers, responding to natural language prompts rather than being trained to win a game. It works alongside human players, undertaking actions based on instructions.

SIMA doesn’t focus on a specific game title but is designed to assist players in achieving their goals. It operates on image input from the environment and user instructions, demonstrating skills in various areas like navigation and object interaction.

Google DeepMind reports that SIMA was trained on nine different video game titles spanning various genres without requiring access to game source code or APIs. The focus was on open-play games to enhance SIMA’s general gaming skills.

While an AI gaming companion like SIMA can enhance multiplayer or offline two-player gaming experiences, there are concerns about its potential misuse. Players could exploit SIMA for unfair advantages in online gameplay, affecting tasks or quests that require human effort.

Google DeepMind acknowledges these concerns and emphasizes that SIMA is a research project aimed at testing AI capabilities in gaming environments. It reassures that there are no immediate plans for public use and measures will be taken to prevent misuse in gaming.

Google clarifies that its AI gaming agent is currently undergoing learning processes and not intended for public release at this time.