Google has also been running a parallel experiment with using AI to remake its core search interface, launching a generative search experience that serves up chatbot-like answers ahead of the familiar list of ads and links.
The company said just a few weeks ago that it doesn’t anticipate a “lightswitch moment” when the generative search experience fully replaces Google Search as we know it. But Google plans to push “the boundaries of what’s possible” and think about “which use cases are helpful and that we have the right balance of latency, quality, and factuality,” Liz Reid, vice president and general manager of Search, said at the time. Like Pichai, she seems to think it’s time to experiment with some radical alternatives to Google’s established model.
Pichai says that Google is focused right now on getting the generative AI experience right, but that he is “open to possibilities around both” paid and ad-supported generative AI experiences. He declines to say whether the paid Gemini offering will remain totally ad-free, but pointed to another Google-owned product where it’s possible to banish ads entirely.
“YouTube has been a very good example of this,” Pichai says, a reference to the paid, ad-free tier that YouTube started experimenting with several years ago. “Ads allows us to give products to more people, but there will be cases of subscriptions that allow people to get a different experience.” He adds, “I can imagine the same user going back and forth between free search and a Gemini subscription.” In other words, generative search would no longer be a side dish to search, but a main menu item—albeit a more expensive one.
Caution Advised
There’s another big reason why Google might want to charge money for its AI services: It helps defray the massive computing costs associated with training and running a large language model.
“We’re able to project forward over our 25 years—if something on day zero costs this month, then what will it cost to perform the same task a year from now, and so on?” Pichai says. “We’ve factored in the efficiencies we’ll gain on the underlying models, and then we price it in a way that we think makes sense.”
Whatever Google’s motivations behind selling subscriptions to a chatbot, the technology it serves up has to work reliably. Pichai acknowledges that Google Gemini, even the advanced version, still risks hallucinating, the way Bard did, or as other generative AI apps have. “We want people to be aware of that,” Pichai says. “I think the technology is useful for many people. But it has to be used in the right way and I still have concerns about people relying on it.”
Pichai says, of course, that Google is trying to reduce the models-gone-wild phenomenon. But he also cautions that the word “hallucinate” should be used carefully, and suggests hallucinating was a feature as well as a bug, which is a fascinating rebranding of misinformation. He believes the technology should be grounded in factualness, but if you dial it down too much, your chatbot gets real boring real fast.
A generative AI experience should be “imaginative,” Pichai says. “Like a child who doesn’t know what the constraints are when they’re imagining something.” Kind of like the early days of the web.