Google is developing an AI based on Deep Mind. Not a winner-take-all opponent or super robot, but a general, instructable gameplay AI agent.

Google published a research paper (via @rowancheung) calling it Scalable Instructable Multiworld Agent (SIMA). The general idea is that it is a learning AI that can follow verbal instructions and understand virtual worlds. Google's SIMA promises to be a more natural and human-like gaming companion, rather than functioning as an opponent or bot for the hard-coded AI we have been accustomed to for decades.

Google has partnered with eight game studios to test the SIMA model in games such as Valheim, Goth Simulator 3, and No Man's Sky. These are open-world games chosen to teach SIMA common game skills; the current version of SIMA can perform about 600 basic skills, including navigation, object interaction, and menu use.

As a player of "No Man's Sky," I am very intrigued by its potential. Anyone who has played the game knows that resource gathering and construction is a tedious process, and telling the AI player to go find oxygen or build a base would really help reduce that grinding aspect. I prefer things like exploration and dogfights to traversing the landscape to find X to build Y to build Z.

Initial results look promising. According to Google, trained SIMA agents performed nearly as well in an unknown game as agents trained specifically for that game. If such agents can help real gamers, that is the key. If a typical game play lasts tens or hundreds of hours, the AI companion must be competent from the opening scene.

While this is an impressive and very interesting feat, I am still skeptical. Large-scale verbal AI models, while adept at acting on knowledge, are limited in their ability to react, adapt, and take appropriate action in real time like humans. How good are they at making headshots on the other side of the map?

AI in video games is not a new concept. Boss fights, multiplayer bots, opponents, NPCs in single-player games, etc., go back decades, but they are hard-coded in every game. Properly implemented, AI like SIMA can have a dramatic impact on general game play. Moreover, it does not require access to source code or an API to function. Instead, it relies on just two inputs: an image on the screen and instructions from the user.

Google emphasizes that this research is still in its early stages. It is expected to become more versatile and adaptive as it learns from more games. Could AI be your next collaborator? It will be interesting to see how this technology develops and what impact it will have on the future of multiplayer gaming.