Google once just wanted to organize the world’s information. Now it seems more determined to feed that information into artificial intelligence algorithms that become obedient, ever-present and increasingly powerful virtual assistants.
Google today announced Gemini 2, a new version of its flagship AI model that has been trained to plan and execute tasks on a user’s computers and on the web, and can chat like a person and make sense of the physical world as a virtual butler.
“I’ve dreamed of a universal digital assistant for a long, long time, as a stepping stone on the road to artificial general intelligence,” Demis Hassabis, CEO of Google DeepMind, told WIRED before the announcement of ‘ today, alluding to the idea of ​​AI that can eventually do anything a human brain can do.
Gemini 2 is primarily a step up in AI intelligence as measured by different benchmarks used to evaluate these things. The model also has enhanced “multi-modal” abilities, meaning it’s more adept at analyzing video and audio, and at speech conversation. The model has also been trained to plan and execute actions on computers.
“Over the past year, we’ve been investing in developing more agent-based models,” Google CEO Sundar Pichai said in a statement today. These models, Pichai added, “can understand more about the world around you, think several steps ahead, and take action on your behalf, with your supervision.”
Tech companies believe so-called AI agents could be the next big leap forward for technology, with chatbots increasingly taking over user tasks. If successful, AI agents could revolutionize personal computing by routinely booking flights, organizing meetings, and analyzing and organizing documents. But getting the technology to reliably follow open orders remains a challenge, with the risk that mistakes can translate into costly and difficult-to-undo mistakes.
Still, Google thinks it’s moving in the right direction and is introducing two specialized AI agents to demonstrate the potential of Gemini 2’s agent: one for coding and one for data science. Instead of auto-completing sections of code, as current AI tools do, these agents can do more complex work, such as checking code into repositories or combining data to enable analysis.
The company is also showing off Project Mariner, an experimental Chrome extension that is capable of taking over web browsing to perform useful tasks for users. WIRED was treated to a live demo at Google’s DeepMind headquarters in London. The agent was asked to help plan a meal, which saw it navigate the Sainsbury’s supermarket chain website, log into a user’s account and add relevant items to their shopping basket. When certain items were unavailable, the model chose suitable replacements based on her own knowledge of cooking. Google declined to perform other tasks, suggesting it remains a work in progress.
