Google's advanced artificial intelligence lab DeepMind has trained a new model that can play high-end games like No Man's Sky, Valheim, and even Goat Simulator without requiring any special instructions.
Named SIMA (Scalable Inscrutable Multiworld Agent), the agent can follow natural language instructions from the player and perform tasks in a variety of game settings.
It builds on years of research teaching AI how to play games, from IBM Watson in chess in the 80s to neural networks that can build games from scratch.
DeepMind researcher Tim Harley says SIMA is not built to win games, but to follow player instructions. Essentially, it's like having your own Gemini AI assistant that can interact with the virtual world.
It opens up the possibility of having a virtual teammate who strives for victory at all costs and can spare you double trouble and do the boring work.
What is SIMA and how was it trained?
Introducing SIMA: the first generalist AI agent to follow natural language instructions in a wide range of 3D virtual environments and video games. 🕹️ Can complete tasks similar to humans, and only he performs better than trained agents in one setting. 🧵 https://t.co/qz3IxzUpto pic.twitter.com/02Q6AkW4uqMarch 13, 2024
SIMA's big advance was independence. While teaching an AI to play one game is a significant accomplishment, the DeepMind team has successfully built a generalist AI agent.
It can work with a variety of games, navigate different environments, and perform tasks from simple natural language instructions.
“This study is the first to demonstrate that an agent can understand a wide range of game worlds and perform tasks within them, following natural language instructions, just as humans can,” the researchers explained. research paper It was released along with a new model.
That training included playing video games and being exposed to as many environments as possible. They worked with developers like Hello Games and Tuxedo Labs to help SIMA understand navigation, menu usage, flight, and crafting.
“By learning from different game worlds, SIMA captures how language is connected to gameplay behavior,” the researchers explained.
What can SIMA do that game AI can't?
SIMA can do anything a human can do in a game, but it is not trained to win. This is very similar to what happens when I try to play a game.
What's really impressive is that SIMA can exist throughout a game without requiring special access to data that is normally unavailable to bots or players.
SIMA can do anything a human can do in a game, but it is not trained to win. This is very similar to what happens when I try to play a game.
DeepMind has built an AI that can learn as you play, without having to be fine-tuned for a specific game or accessing the game's source code to accomplish a prescribed goal.
Version 1 is trained in 600 basic skills, including navigation, object manipulation, and menu use. You can complete any task in less than 10 seconds, and it's usually much faster.
“We want future agents to work on tasks that require advanced strategic planning and multiple subtasks to complete,” the team explained.
This could include a human player asking a SIMA-equipped teammate to go somewhere to find resources to build a camp, or walking around an existing camp when an enemy approaches. may include sounding an alarm.
Introducing a new type of gameplay
Bots in games aren't new, and there's no AI as companions, guides, or even ways to simplify game play on easy mode. But this takes the concept to a new level.
In the future, a personal assistant built on SIMA technology could sit alongside your Steam, Xbox, or Playstation account and allow you to participate in supported games.
Personally, I would welcome something to fend off attackers when participating in multiplayer games. In Rust it may last more than 2 minutes before being shot.
