Google unveils AI agent capable of autonomous web navigation

Google has unveiled its latest innovation in artificial intelligence, a prototype AI agent named Mariner, designed to autonomously perform tasks like navigating websites and using online tools. Mariner represents a significant leap in AI technology, showcasing Google's ambitions in creating versatile digital assistants. The AI agent is powered by Gemini 2.0, Google’s advanced neural network system, which underpins many of the company’s AI-driven products and research.

Gemini 2.0, described as a neural network, learns from vast datasets encompassing text, images, and sounds. This allows it to recognize patterns and take actions based on user needs. According to Demis Hassabis, head of Google’s core AI lab, Mariner can “take action in the world” by understanding how to interact with digital environments. For example, it can fill a virtual shopping cart or manage a spreadsheet. However, actions like making purchases remain under user control, ensuring human oversight.

Jaclyn Konzelmann, a project manager at Google, emphasized that Mariner operates “with a human in the loop.” This collaborative approach prioritizes user guidance while allowing the AI to handle repetitive or complex tasks. Despite its capabilities, Mariner remains experimental. Konzelmann acknowledged that the system, like other AI models, is prone to errors and is not yet fully accurate.

Currently, Mariner is being tested by a limited group of external users. Google has not announced a timeline for wider availability. Sundar Pichai, Google’s CEO, highlighted in a blog post that these advancements are a step closer to realizing the company’s vision of a universal AI assistant.

In addition to Mariner, Google introduced an updated version of Project Astra, a digital assistant that responds to images, text, and verbal commands. Astra, like Mariner, is not yet available to the public and remains in the experimental phase.