All this will be made possible thanks to ever more omniscient assistants. Google presented, on Tuesday May 14, its latest innovations in generative artificial intelligence (AI), which will potentially transform the daily lives of its users, from online research to many everyday tasks.
“The most exciting transformation is obviously generative online search on Google,” said boss Sundar Pichai, on stage at the company’s open-air amphitheater in Mountain View (California).
For a year, Google has been testing its search engine powered by generative AI, that is to say the production of content on a simple query in everyday language. The Internet user enters their query and receives, at the top of the results page, a response generated by Gemini, Google’s AI model. He can then click on suggested questions, or, further down, on links to websites.
With these new options, the tech giant hopes to make it easier for users, by helping them concretely compose their menus or plan their vacations. “Users are doing more research and are more satisfied,” assures Sundar Pichai.
The new formula – the most significant transformation of the search engine since its creation – will be deployed in the United States this week, then in other countries, to reach more than a billion people by the end of 2024. Google dominates the online search to the point that its name is synonymous with the action. But the eruption of ChatGPT and other chatbots capable of answering users’ questions threatens its empire.
Research under threat –
All of Silicon Valley is competing for new tools and virtual contacts to circumvent the world number one in digital advertising. On Facebook, Instagram and WhatsApp, for example, users can ask questions to Meta AI, which has access to the internet. As a result, Gartner predicts that by 2026, the volume of queries to traditional search engines will drop by 25%.
The battle therefore shifts to AI assistants, which seem to gain new powers every week, thanks to advances in generative AI models.
Google’s central model, Gemini 1.5 Pro, can thus integrate an increasing volume of context information provided by the user (reports of hundreds of pages, longer videos, etc.) and digest different formats (the model also “understands” text, sound and images, and can respond in writing, by voice or by generating images).
Google DeepMind, the group’s research laboratory, for its part presented other new models: Gemini 1.5 Flash (faster and less expensive), Imagen 3 (image generation) and Veo (video generation, a growing sector). booming). In the long term, Sundar Pichai envisions AI agents as “intelligent systems capable of reasoning, planning and retaining information. They anticipate steps and know how to work with software to accomplish things on your behalf, under your supervision”.
Custom wizards
With nearly $74 billion in profits in 2023, and numerous services omnipresent on a daily basis (Google, Gmail, Maps, Android, YouTube, etc.), the company is ideally placed to bring this vision to life. In the Gemini mobile app, users can chat with the model “like a friend” and will soon be able to create “Gems”, personalized assistants, like a personal sous chef or yoga instructor .
Gemini is also integrated directly into Android, Google’s mobile operating system. The model can thus intervene at any time to anticipate the person’s needs – during a call, to report potential fraud, or in messaging to compose responses. And the advances of Project Astra, an AI agent prototype, should enrich these experiences.
Google DeepMind broadcast a video demonstration: we see a user pointing the camera of their smartphone – or glasses with an integrated camera – at their environment, and interrogating the model which identifies the place where they are, solves a computer problem based on of a diagram and remembers where he placed an object.
On Monday, OpenAI, the startup that launched the generative AI revolution, delivered a similar presentation, where ChatGPT interacts with an engineer in a way so natural that the machine appears human. But the fluidity of conversation does not yet make chatbots omniscient, proactive and personalized AI agents.