Google will use artificial intelligence to return summarized responses to search engine queries from US users as it continues to infuse generative AI into its most widely used products.
The company has been testing “AI overviews” that appear at the tops of search results, summaries created by its Gemini AI model that appear alongside the traditional link-based search results.
The featured has also been tested in the UK but will be rolled out across the US beginning on Tuesday, Google announced at its annual I/O developer conference Tuesday in California. Google Search head Liz Reid said AI Overviews would become available to “more than a billion people” by the end of the year.
Google also announced a text-to-video artificial intelligence model called Veo, allowing for the creation of computer-generated footage based only on written prompts. The model is a clear rival to OpenAI’s Sora, which performs similar functions and is planned to be released to the public later this year.
Google additionally revealed a new AI assistant in progress under the working name Project Astra, previewing an early version of the voice tool that can use a smartphone’s camera to verbally identify locations, read and explain computer code and create alliterative sentences.
The assistant was previewed to journalists in a video demo and showed the tool interacting by voice with a Google employee, using the camera lens to identify the view from a window – the London area of King’s Cross, where Google’s AI unit is based – and comprehend computer code. Sundar Pichai, the CEO of Google, said the tool was “pretty magical”, with Google aiming for its debut in the second half of this year.
The company said its overviews feature would be able to handle complex questions such as finding Pilates studios in the US city of Boston, showing their best prices and their walking times from a specific location – all asked in one query. Using Gemini’s technology, Google will start the response with an AI-generated summary, which then links out to other content including web links.
The Veo demonstration showed filmmaker and actor Donald Glover praising the model as AI generations like a sailboat gliding across the sea played onscreen. “Everybody is gonna be a director,” Glover said. “And everybody should be.”
Concerns over AI-generated footage replacing the work of filmmakers and entertainment industry workers has been a major labor rights issue in recent years. Some of the country’s most prominent entertainment unions, such as Sag-Aftra, have gone on strike over issues that include how studios are allowed to use AI. After seeing OpenAI’s Sora earlier this year, filmmaker and studio owner Tyler Perry also announced that he was putting an $800m studio expansion on hold.
The updates were announced a day after OpenAI, the developer of the ChatGPT chatbot, announced a new model called GPT-4o that can interact with people by voice. The model will be a free version of OpenAI’s GPT-4 AI model, which was previously a paid-for product.
The AI overview, which the company described as “taking more of the legwork out of searching”, will appear when Google’s systems decide that such a response can be helpful – for instance, when replying to query that requires collating information from a range of sources. Google indicated that the overview feature will roll out to other countries over the next few months, saying it will be brought to “over a billion people by the end of the year”.
Google said the traditional search format still seemed to benefit from the AI-generated approach, with web links proving even more popular in tests.
“We see that the links included in AI Overviews get more clicks than if the page had appeared as a traditional web listing for that query,” said the company. Amid concern from publishers that news-related queries will be met with AI-generated responses and not links to original journalism, Google said it would “continue to focus on sending valuable traffic to publishers and creators”.
Google also announced a faster version of its Gemini model, called 1.5 Flash, and an updated version of its image-generator model. This year, Pichai said some Gemini-made images were “biased” and “completely unacceptable” after the model produced results including portrayals of German soldiers in the second world war as people of colour.