Google is working on an AI called SIMA, which aims to be the perfect co-op partner for you.
Everyone is into AI right now, including tech giant Google. With Google DeepMind, the company is exploring multiple avenues for how AI can be used in the future, and video games are obviously a big potential space for that.
After introducing an AI that “creates” games within seconds (albeit in terrible quality), Google has now revealed another interesting video game-related AI with The Scalable, Instructable, Multiworld Agent, or short: SIMA. This AI is set to be your next co-op partner, and hopefully this one won't mess up as badly as your current ones always seem to do.
Google AI SIMA Just Wants To Be Your Friend
The goal of the SIMA project is “to develop an instructable agent that can accomplish anything a human can do in any simulated 3D environment”. Or, in layman’s terms, turn the AI into a gamer. The purpose and progress of SIMA have been detailed in a new research paper, which is well worth a read. The core idea is to improve dynamic reactiveness of AI, which is currently one of the biggest issues in things like Large Language Modules, or LLMs. ChatGPT can recite Wikipedia quite well, but anything beyond that can pose issues.
SIMA is set to go far beyond what we have always referred to as AI in video games and is supposed to offer much more dynamic results and behavior that is closer to how humans interact with games. To train the AI, Google DeepMind has collaborated with several video game developers to train SIMA on their games, which includes titles such as No Man’s Sky, Teardown, Valhalla and Goat Simulator 3. Yep, odd choice that one!
The AI was tested on 600 basic skills, some results of which you can see on the official website. This includes finding resources, opening certain menus or jumping over an obstacle. It’s definitely far from perfect, but you can already see the potential in this.
As rightfully skeptical as one should be of AI and its current developments, I wouldn’t mind having AI in games be much more advanced than it is nowadays. I love playing co-op games, but coordinating with other people can be hassle. And hey, so many games have annoying tasks in them that you’d rather send a robot to do, right? Having a fully functioning, dynamic and reactive AI companion by your side, who can basically do everything you can, could be a really cool addition to our co-op experiences.
What do you think of Google DeepMind’s AI experiment? Do you see potential in this or is it not that exciting? Let us know in the comments!