Google, which was once focused solely on organizing the world's information, is now concentrating on integrating this information into AI algorithms to create powerful virtual assistants. The company announced Gemini 2, an advanced AI model designed to perform tasks on computers and the web, chat like a person, and function as a virtual butler. According to Demis Hassabis, CEO of Google DeepMind, Gemini 2 represents a step toward creating a universal digital assistant and eventually artificial general intelligence. Gemini 2 showcases heightened abilities in processing video and audio, understanding speech, and executing actions on computers. Google's CEO, Sundar Pichai, noted the company's investment in developing "agentic models" that can comprehend the world, plan, and act under user supervision. AI agents, which could revolutionize personal computing, might soon automate tasks like booking flights and organizing documents, although ensuring their accuracy remains challenging. Google has unveiled AI agents for coding and data science, which can perform complex tasks beyond current AI tools. The company also introduced Project Mariner, an experimental Chrome extension that can handle web navigation for users.

In a demonstration, Mariner was tasked with planning a meal and successfully navigated to a supermarket website to add items to a cart, while making appropriate substitutions. Meanwhile, the Gemini 2 model, launched to catch up with OpenAI's ChatGPT, offers capabilities in line with ChatGPT and integrates AI into Google's search products. By understanding audio and video, Gemini 2 aims to transform interactions with digital assistants. An experimental project called Astra allows Gemini 2 to interpret its surroundings through a camera and converse naturally based on its observations. WIRED tested Astra's capabilities, which included offering wine recommendations and art history insights by referencing web sources. As Gemini 2 learns users' tastes and interests, concerns about privacy and security persist. Hassabis emphasizes the importance of understanding user behavior and addressing privacy issues from the start. While Gemini 2 has demonstrated impressive adaptability, acknowledging its potential for errors and the need for ongoing refinement is crucial.