Google Introduces SIMA, AI that can Play Games like Humans

Google DeepMind has launched SIMA, an artificial intelligence agent designed to play video games like a human. SIMA is trained to interpret natural language instructions and navigate three-dimensional environments in games.

Contents

Skills Acquired by SIMA Training Method Training Approach and Interface of SIMA Future Prospects and Goals

Please follow us on Twitter and Facebook.

Reasearchers at Google DeepMind explained that the primary objective behind developing SIMA is to create an AI assistant for human players in video games. Unlike traditional AI, SIMA is not focused on achieving high scores but on understanding and fulfilling objectives based on natural language instructions.

SIMA serves as an assistant for human players, responding to specific requests and completing tasks within the game environment. It is not intended to replace existing AI in games but to enhance the gaming experience by providing assistance to players.

Skills Acquired by SIMA

Through training with various video games, SIMA has acquired approximately 600 basic skills. These skills include navigating environments, interacting with objects, and completing simple tasks within the games.

Training Method

Google used a unique training method for SIMA, which involved studying player pairs and recording gaming sessions. Additionally, DeepMind designed a dedicated environment for AI training, focusing on the manipulation of objects using simple blocks.

Google SIMA by DeepMind — Source: Google DeepMind

Training Approach and Interface of SIMA

SIMA’s training does not require custom APIs or access to source code. It utilizes on-screen images and natural language instructions provided by the user. This simple interface allows SIMA to potentially interact with any virtual environment. This is how Californians explain it:

“[The agent] requires only two input methods: on-screen images and simple natural language instructions provided by the user. SIMA uses keyboard and mouse commands to control the central character of the games to carry out these instructions.’ This simple interface is what humans use, meaning SIMA can potentially interact with any virtual environment.”

Future Prospects and Goals

Google aims for SIMA to become more versatile with additional training, potentially extending its usefulness beyond gaming. Ultimately, the research is directed towards developing more general AI systems that can understand and execute a wide range of tasks online and in the real world. Californians also stated:

“Ultimately, our research is directed toward more general AI systems and agents that can safely understand and carry out a wide range of tasks in a way that is useful to people online and in the real world.”