“Project Jarvis”, Google’s AI that takes over your computer (and buys for you)

The roaring entry on the scene of artificial intelligence (AI) is freeing flesh-and-blood human beings from a plethora of tasks (repetitive and tedious) that can be perfectly assumed by fashionable technology. But to what extent would we be willing to entrust AI with potentially sensitive tasks (if some type of error occurs) such as purchasing a flight? Google would be developing an AI named “Project Jarvis” that would be able to take control of the browser on the user’s computer to execute all kinds of tasks there on their behalf.from compiling information, purchasing a product or booking a flight. This is how he publishes it The Information based on the two provided by three people who have direct knowledge of this project of the Mountain View company.

Powered by an upgraded version of Gemini, “Project Jarvis” would be an AI with the focus specifically on the user’s browser (Chrome to be specific). And its goal is to help the user “automate everyday activities that are at the mercy of the web” by taking and interpreting screenshots and then clicking on buttons or writing text, he says. The Information. In its current phase of development, Google’s new AI would take just “a few seconds” to complete a task and then move on to the next.

Google’s AI rivals are also working on similar models.

The most prominent companies in the AI ​​space are all working on models that operate similarly to “Project Jarvis.”. Microsoft’s Copilot Vision can, for example, engage in conversations with the user about the web pages they are viewing. Apple Intelligence will also rely on what appears on the user’s device screen to multitask across apps (and this feature will debut sometime next year). Similarly, Anthropic recently presented an update to the Claude chatbot that can use the computer on behalf of the user, while OpenAI is also working on a similar model.

Initially, Google would intend to present “Project Jarvis” next December. However, the company would be considering testing this model first among a small number of users to detect any errors and correct them before a more massive launch of this technology.