By Jeffrey Dastin and Greg Bensinger
MOUNTAIN VIEW, California (Reuters) -Alphabet Inc’s Google is rolling out more artificial intelligence for its core search product, hoping to create some of the same consumer excitement generated by Microsoft Corp’s update to rival search engine Bing in recent months.
At its annual I/O conference in Mountain View, California, on Wednesday, Google offered a new version of its namesake engine. Called the Search Generative Experience, the revamped Google can craft responses to open-ended queries while retaining its recognizable list of links to the Web.
“We are reimagining all of our core products, including search,” Sundar Pichai, Alphabet’s CEO, said after he took the stage at the event.
He said Google is integrating generative AI into search and other products, including Gmail, where it can create draft messages, and Google Photos, where it can make changes to images like centering figures and coloring in empty space.
Alphabet’s shares were up nearly 4%.
U.S. consumers will gain access to the Search Generative Experience in the coming weeks via a wait list, a trial phase during which Google will monitor the quality, speed and cost of search results, Vice President Cathy Edwards said in an interview.
Google’s foray into what is known as generative AI comes after the startup OpenAI introduced ChatGPT, the darling chatbot of Silicon Valley that launched a furious funding race among would-be competitors. Generative AI can, using past data, create brand new content like fully formed text, images and software code.
OpenAI, backed by billions of dollars from Microsoft and now integrated into Bing search, has become for many the default version of generative AI, helping users spin up term papers, contracts, travel itineraries, even entire novels.
For years the top portal to the internet, Google has found its own perch in question since rivals began exploiting the technology as an alternative to presenting content from the Web. At stake is Google’s slice of the gigantic online advertising pie that the research firm MAGNA estimated at $286 billion this year.
Edwards told Reuters that Google’s aim was to employ generative AI to reduce steps for consumers to make decisions and to let people ask a broader set of queries, including creative ones. Addressing how AI can spout incorrect information, Edwards said the company prioritized accuracy and citing trusted sources.
“AI can provide insight,” Edwards said. “But what fundamentally people want at the end of the day is to be connected to information from real people and organizations, knowing, for example, that this health information comes from the WHO,” or the World Health Organization.
WHAT OUTFIT TO WEAR
With the embedded AI, Google still looks and acts like its familiar empty search bar.
But while a search for “weather San Francisco” will as usual point a user to an eight-day forecast, a query asking what outfit to wear in the California city prompts a lengthy response generated by AI, according to a demonstration for Reuters earlier this week.
“You should bring layers, including a short-sleeved shirt and a light sweater or jacket for the day,” the result stated, including links to websites where it gleaned such advice.
Searches for news or to navigate to a specific website did not result in a response by generative AI, though users can prompt a chat and press a button for a follow-up query, the demonstration showed.
Bing search and Microsoft’s corresponding Edge browser, through its partnership with OpenAI, can summarize Web pages, synthesize disparate sources, compose emails and translate them. Microsoft has said every percentage point of share it gains in search advertising could draw another $2 billion in revenue.
Another challenge is the high expense of drawing on such AI, known as large language models. Edwards said, “We and others are working on a variety of different ways to bring down the cost over time.”
Ads will remain a central part of the experience, Edwards said. “We only get paid when there’s a click.”
BARD FOR 180 COUNTRIES
In more news for I/O, Google Cloud CEO Thomas Kurian told Reuters the division is lining up customers to put its newest technology to the test, among them Deutsche Bank AG and Uber Technologies Inc.
In recent years, however, Google’s rivals have taken its research breakthroughs and run with them in products, outpacing their inventor.
ChatGPT came to light after an AI system Google revealed in 2017. The speed at which the chatbot grew – faster than any consumer application in history – encouraged the often deliberative Google to prod staff to hurry along projects so they were ready for public consumption.
In February Google announced its competing chatbot called Bard. A promotional video that month that showed Bard answering a question incorrectly propelled a stock slide shaving $100 billion off Google’s market value.
Now, Bard will be multimodal like OpenAI’s GPT-4, the company said on Wednesday, and it will make the chatbot accessible to people in more than 180 countries and territories.
That means customers will be able to prompt Bard with images, not just text – for instance asking the chatbot to write a caption to a picture they hand it, it said.
Behind Bard also is a more powerful AI model Google announced called PaLM 2, which it said could solve tougher problems. One of its PaLM 2 models is lightweight enough to work on smartphones, Pichai also said.
(Reporting by Jeffrey Dastin and Greg Bensinger in Mountain View, CaliforniaAdditional reporting by Yuvraj Malik in Bengaluru and Sheila Dang in DallasEditing by Sayantani Ghosh, Peter Henderson and Matthew Lewis)
