New Google AI teaches browsing in Chrome

New Google AI teaches browsing in Chrome

Google is working on a new AI system that can browse Google Chrome like a human. Among other things, the AI will be able to fill out forms, and understand screenshots.

Google is building an AI model that can do browsing in Google Chrome. The model is internally named Project Jarvis, and should be able to automate tasks such as shopping, research and booking flights. Google is using its own Gemini 2.0-LLM as a base. That’s what The Information knows.

Project Jarvis will be able to read websites, interpret screenshots, press buttons and enter text. Basically, the AI should be able to surf the Web via Chrome like a human. Currently, though, the system would need several seconds of time each time between different actions.

Autonomous RPA bot

In this way, Project Jarvis sounds like a kind of RPA bot for Web pages, which can work autonomously. RPA automates repetitive operations through the UI built for humans, while Jarvis has to do unique work through the UI.

Autonomous capabilities for AI are becoming increasingly important. Whereas last year the copilot story was still central, major AI specialists now want to develop models that can really take over certain tasks from humans with confidence. In that regard, Anthropic last week showed a new version of Claude that can work independently with a computer. By comparison, Google’s ambition is relatively limited, as Project Jarvis is currently focused only on the browser.

Google itself remains sparse with details about Project Jarvis. The model would be introduced to the general public in December.