What is Google Deepmind's new AI SIMA that can play video games with you | News explained

and AI robot overcomes loneliness in the elderly and Fully autonomous AI software engineer It's been a busy week for AI enthusiasts with recent releases. Here's what you need to know about SIMA:

What is Shima?

AI research organization Google Deepmind describes SIMA as an AI agent, which is different from AI models such as OpenAI's ChatGPT and Google Gemini. AI models are trained on huge datasets, so they have limitations when it comes to working alone. AI agents, on the other hand, can process data and take actions on their own.

SIMA can be described as a generalist AI agent that can perform different types of tasks. It's like a virtual sidekick that can understand and follow instructions in all kinds of virtual environments, from exploring mysterious dungeons to building luxurious castles. Accomplish tasks or solve assigned challenges.

It's essentially a super-smart computer program that you can think of as a digital explorer, with features that understand what you want and help you create it in the virtual world.

How does SIMA work?

SIMA is trained to process human language, so it “understands” your commands. So when you ask it to build a castle or find a treasure chest, it understands exactly what those commands mean.

One of the defining characteristics of this AI agent is its ability to learn and adapt. SIMA does this through interaction with users. The more you interact with SIMA, the smarter it will learn from its experiences and improve over time. This allows us to better understand and fulfill your requests.

Based on the current stage of AI development, it is a huge feat for an AI system to be able to play even a single game. However, SIMA goes beyond that and allows you to follow instructions in a variety of game settings. This could lead to the introduction of more useful AI agents in other environments.

Google DeepMind's latest research shows that the functionality of advanced AI models can be translated into “practical real-world actions through language interfaces.” With this, Google hopes that his SIMA and other AI agents will be able to use video games as a sandbox to understand how AI systems can help them.

How was SIMA trained?

The company's official blog explains: “We partnered with a game developer to train his SIMA on a variety of video games. This research shows that the agent, like humans, understands a wide range of game worlds and navigates them by following natural language instructions. This is the first demonstration that it can perform tasks with

To expose AI agents to different environments, Google Deepmind worked with eight game studios to test SIMA on nine different video games, including Tuxedo Labs' Teardown and Hello Games' No Man's Sky.

With each game, SIMA's portfolio has opened up new interactive worlds with different skills to learn, such as simple navigation, using menus, mining resources, and flying spaceships.

We also used four research environments, including one built in the cross-platform game engine Unity. The environment, called Construction Lab, required AI agents to construct sculptures from building blocks. This exercise was done to test the agent's ability to manipulate objects and intuitive understanding of the physical world.

By learning from different game worlds, SIMA understands how language is connected to gameplay behavior, according to the researchers. “Our initial approach was to record pairs of human players across games in a portfolio, with one player monitoring and directing the other. We also gave players the freedom to play. “We then asked them to review the player's actions and record the instructions that lead to the game's actions,” the official blog reads.