Google Just Added Generative AI to Search
SIX MONTHS AGO, Google didn’t appear remotely worried about its search business. Then OpenAI’s ChatGPT was unleashed, and Microsoft’s Bing spawned a chatbot.
At Google’s annual I/O conference today, the search giant announced that it will infuse results with generative artificial intelligence technology similar to that behind ChatGPT. The company is launching an experimental version of its prized search engine that incorporates text generation like that powering ChatGPT and other advanced chatbots.
Google’s reimagined search still involves typing a query, and it still responds with links to websites, snippets of content, and ads. But in some situations, the top of the page will feature text synthesized by AI that pulls from information found on different sources across the web, and link to those webpages. A user can ask follow up questions to get more specific information.
A query about the coronation of Britain’s new king might be met with a couple of paragraphs summarizing the event. If asked about ebikes, Google’s algorithms can list bullet-point recaps of product reviews published by various websites, and link to online stores where a user can make a purchase. The revamped version of search will be accessible in the US via a new feature called Search Labs, but it will not be activated by default or for all Google users.
Google’s AI-infused search is considerably tamer than ChatGPT, eschewing an anthropomorphized persona and avoiding topics that might be deemed controversial, such as politics and medical or financial advice. When asked if Joe Biden is a good president or for information about different US states’ abortion laws, for example, Google’s generative AI product declined to answer.
“The technology is very early on, it has its challenges, and we will make mistakes certainly,” says Liz Reid, vice president of search at Google, who gave WIRED a preview of the new features ahead of I/O.
Google is moving quickly to add ChatGPT-like features to search, but whether users will find them useful remains to be seen. Product searches, for instance, synthesized material from different reviews, but it was not immediately obvious how the brief summaries might improve the search experience.
The unpolished feel of these new features may reflect the fact that their launch is a defensive move. Google has invested huge sums and major resources in AI over recent years, with CEO Sundar Pichai often calling the company “AI first.” Yet Google still found itself wrong-footed with the arrival of ChatGPT, a surprisingly clever and garrulous—though also fundamentally flawed—chatbot from OpenAI.
ChatGPT is powered by a machine learning model trained to predict the words likely to follow a string of text by digesting huge amounts of text, including vast numbers of web pages. Additional training, provided by humans rating the quality of the bot’s responses, made ChatGPT more adept at answering questions and holding a conversation.
Because ChatGPT was trained on much of the web, users quickly found it a promising new way to search, even if the bot is prone to fabricating information. Microsoft seized on this potential by investing $10 billion in OpenAI in January and then incorporating ChatGPT into Bing a month later.
The hype and strong interest from users, who fed queries to ChatGPT and Bing chat by the millions, left Google scrambling to catch up. Researchers at the company had developed some of the core technology at work in the new chatbots, but Google had been cautious about publicly launching its precursor to ChatGPT, called LaMDA.
In March, Google changed strategy, announcing a ChatGPT competitor called Bard. In April, Google said it would combine its AI research group with another Alphabet company focused on AI, DeepMind. Now Google is infusing text-generation technology into its core product, search.
Google’s dominance and reputation could make this latest move the biggest test yet of the power and usefulness of ChatGPT-style technology. It’s also risky.
Because language models can fabricate, companies have to develop ways to check that the information served to users is accurate. And some online publishers are concerned about search companies scraping and regurgitating their content in ways that mean fewer referrals. Google also needs to avoid cannibalizing its search advertising business, which provides a significant chunk of the company’s revenue.
Google’s cautious rollout “suggests they are not ready for any major changes to search yet,” says Aravind Srinivas, CEO of Perplexity AI, a startup also working on generative AI for web search. He says the way Google is combining generative content with conventional search results—not replacing them—shows how hesitant the company is hesitant to mess with its search advertising business.
The capabilities of ChatGPT and other advanced AI models have done more than just unsettle Google. Some experts worry that they signal accelerated progress in AI, which might lead to technology that’s more difficult to contain and control. Geoffrey Hinton, a pioneer in the field and formerly one of Google’s most respected researchers, recently left the company in order to draw attention to the dangers he believes these AI models raise.
At I/O, Google is announcing a dizzying number of new projects and services that use the kind of generative AI found in ChatGPT. But its first efforts to fold the technology into a dominant search product may be the most consequential for its many users, and for the company itself.
Knight, W. (2023, May 10). Google I/O 2023: Google Adds Generative AI to Search. WIRED. https://www.wired.com/story/google-io-just-added-generative-ai-to-search/