Updated: March 15, 2024 12:59 PM EDT This part has been updated to include additional information provided to Mashable by Google DeepMind.
Computer-controlled video game characters are nothing new. Non-playable characters or his NPCs are a long-standing staple of gaming.
But Google's AI research lab, DeepMind, announced We're working on something unique and new in computer-controlled gameplay: SIMA, which stands for Scalable Instructable Multiworld Agent.
Tweet may have been deleted
Unlike NPCs, Google DeepMind's AI-controlled game companions are not programmed into a specific game to behave in a particular way or respond to specific human actions. As DeepMind describes it, SIMA is a “generalist AI agent for 3D virtual environments.”
SIMA is not an NPC
SIMA plays alongside the gamer as a human player, following verbal commands. Google explains that “SIMA is not trained to win games.” It is prepared to cooperate with human players and perform actions based on natural language prompts.
This element is key to what makes SIMA unique. Again, SIMA is not an NPC specifically designed for a particular title, intended to provide a challenge to the player. SIMA acts as a fellow gamer trying to accomplish whatever the human player tells it to do.
“SIMA only requires images provided by a 3D environment and natural language instructions provided by the user,” Google said. “It uses mouse and keyboard output to be assessed across 600 skills across areas such as navigation and object interaction (such as ‘turning left’ and ‘chopping down a tree’).”
According to Google DeepMind, SIMA was trained and tested on nine different video game titles from various game genres. no man's sky, goat simulator 3, tear downand A shaky life. This training does not require any access to the game's source code or API. Google Deep Mind To tell To facilitate the training of SIMA's more general gaming skills, we focused on open play games rather than narrative games.
SIMA has some good points and some obvious bad points. AI Game Companion is useful if you want to play multiplayer games or offline his 2-player story mode without another human selecting a controller.
But the ideas driving SIMA's research can also clearly be used for malicious purposes. A gamer may use SIMA to gain an unfair advantage in online gameplay. Games that require players to perform tedious tasks for resources or complete quests to earn XP can be operated by a human simply instructing her AI companion to do all that work. It is clear that it can be done.
Google DeepMind reached out to Mashable to share that they have similar concerns about unintended uses of SIMA, as many on the team are gamers or former game developers.
DeepMind reiterated that SIMA is a research project and the video game is a sandbox the team is using to test the capabilities of its AI agents. DeepMind says it currently has no plans to make SIMA widely available and intends to keep it from impacting the bot problem in gaming.
As shared by Google, AI agents that play video games are not ready for public use at this time. For now, I'm still learning how to play new video games.
