Artificial intelligence (AI) is being incorporated into many apps and services, and video games aren"t untouched by it. Last year, Nvidia announced a new technology that lets developers create generative AI-powered NPCs that can talk to player characters in real time. Google DeepMind"s new AI takes things a step further by learning human-like gaming skills.
Google"s DeepMind research group unveiled a new AI program that is capable of learning how to complete tasks in a number of games. Called Scalable Instructable Multiworld Agent (SIMA), the AI program can follow natural-language instructions to carry out tasks in a variety of video game settings.
Google DeepMind stated:
SIMA is an AI agent that can perceive and understand a variety of environments, then take actions to achieve an instructed goal. It comprises a model designed for precise image-language mapping and a video model that predicts what will happen next on-screen. We finetuned these models on training data specific to the 3D settings in the SIMA portfolio.
Google teamed up with eight game developers to train and test SIMA, including big names like Hello Games and Coffee Stain. To teach SIMA the ropes of gaming, the researchers plugged it into titles like No Man"s Sky, Teardown, Valheim, and even the Goat Simulator 3.
Google said SIMA "doesn"t need access to a game"s source code, nor bespoke APIs. It requires just two inputs: the images on the screen, and simple, natural-language instructions provided by the user."
SIMA currently has around 600 basic skills, such as turning left, climbing a ladder, and opening the menu to use a map. However, it can"t perform more strategic tasks like finding resources and building camps yet.
The AI program is still in the early stages of research. Google hasn"t revealed whether it will be available for general use anytime soon.