Google’s vision for Gemini involves turning it into a ‘World Model’
What you need to know
- During I/O 2025, Google’s DeepMind CEO, Demis Hassabis, discussed the company’s ambitious outlook for Gemini.
- Google reportedly wants Gemini to become a “universal AI” and morph it into a “world model,” capable of understanding and simulating the world to make plans.
- Hassabis adds that Google wants to incorporate Project Mariner’s multitasking power and Project Astra’s visual understanding to Gemini to help achieve this.
Google’s AI just seems to progress; however, the company is now detailing its true vision for its software.
In a post written by Google’s DeepMind CEO, Demis Hassabis, the company discussed how it wants to go about building a “universal AI.” This plan is firmly set on the Gemini 2.5 Pro model, which Google says is wants to morph into a “world model.” Hassabis states there’s are hopes for Gemini to become an AI that can “make plans and imagine new experiences.” This will reportedly be done by enabling the AI to “understand and simulate aspects of the world.”
Google attributes this way of thinking and forethought to that of the human brain.
Hassabis says the company isn’t starting from zero with these ambitions. The company has reportedly observed instances of this world understanding and simulation in Gemini when natural environments are involved. Google stresses that it wants Gemini to become intelligent, one that can understand the “context you’re in” and “take action on your behalf, across devices.”
Speaking of tasks, Hassabis explains that Google has been mulling over the thought of bringing Project Mariner into Gemini for multitasking.
Project Mariner & Astra
Project Mariner revealed itself to users in December, and Google says it’s been hard at work, working with testers to gather feedback on it. Since December, Mariner has grown to have the ability to complete ten tasks at once, says Hassabis. The agents within Project Mariner enable the software to research information, book shows or other events, and research topics all at once.
It’s this level of multitasking that Hassabis says will help usher Gemini into that “world model” ideal.
What’s more, Project Astra will work in tandem with Mariner to bring Gemini into a more universal setting. Google states Astra’s functions, like video understanding, screen sharing, and memory, will be a great asset. Project Astra has since worked its way into Gemini Live. From that, Google’s taken in a load of feedback from users to bring even more to Gemini Live, Search, and the Live API.
In March, Google announced that it was bringing Project Astra to Gemini. With these new details, we’re now understanding a little better why Google did this, and it’s outlook/future plans for Gemini. At the time, Google said it was bringing Project Astra over from its DeepMind team as it prepares for its public launch.
Post Comment