Ever since AI tools gained popularity, companies have been testing and experimenting with the technology in different areas. While we have AI bots for writing content, generating images, editing videos and so on, there was still no AI chatbot that we could play games with. But now, that might change. Google has just revealed that it is working on SIMA- an AI agent that is being trained to play games like a human professional. The AI agent is being trained in various gaming environments and is expected to be as skilled as a human gamer. Imagine playing a game like Call Of Duty with your friends, but one of these friends is an AI agent who is as skilled as a professional human gamer. SIMA could prove to be immensely useful for gamers as they can test their skills anytime by playing with the AI chatbot.
SIMA stands for Scalable Instructable Multiworld Agent and has been announced by Google in a blog post. The AI agent is capable of understanding natural-language instructions and completing tasks across a variety of video game environments. Google DeepMind, renowned for its advancements in AI, has a history of exploring AI’s capabilities in gaming, from Atari games to StarCraft II.
The blog post also says that video games serve as excellent training grounds for AI systems due to their dynamic and interactive nature.
Google also said in the post that SIMA represents a significant advancement, shifting focus from individual games to a more general approach in game-playing AI. It marks the first instance where an AI agent can comprehend diverse gaming worlds and execute tasks based on natural-language instructions, akin to human players.
“This research marks the first time an agent has demonstrated it can understand a broad range of gaming worlds, and follow natural-language instructions to carry out tasks within them, as a human might,” the company noted.
To train SIMA, Google collaborated with eight game studios and utilised different video games including popular titles like No Man’s Sky and Teardown. Each game presents unique challenges, from navigation to resource management, providing a comprehensive learning experience for the AI.
Unlike previous approaches that focused on single games, SIMA is being trained across multiple environments. Google notes that the AI agent demonstrated remarkable adaptability, performing well even in unseen games, showcasing its potential for generalisation.
SIMA operates through a simple interface, requiring only screen images and natural-language instructions. This accessibility enables it to interact with various virtual environments without needing access to game source code or specialised APIs.
“Our AI agent doesn’t need access to a game’s source code, nor bespoke APIs. It requires just two inputs: the images on screen, and simple, natural-language instructions provided by the user. SIMA uses keyboard and mouse outputs to control the games’ central character to carry out these instructions. This simple interface is what humans use, meaning SIMA can potentially interact with any virtual environment,” the company noted in its post.
The current version of SIMA excels in performing basic tasks across 600 skills, such as navigation and object interaction. Google aims to further enhance SIMA’s capabilities to tackle complex tasks, including strategic planning and multi-step objectives.
Source link
credite