Like it or not, Google is doubling down on AI-generated search.
On Wednesday, the tech giant announced that it's expanding AI overviews to more Google Search queries, starting with advanced math, coding, and multimodal searches. That's made possible due to Google's more advanced model, Gemini 2.0,, which now powers AI overviews.
AI Overviews is also expanding access to more users outside of the U.S. by allowing people who aren't logged in to see the AI-generated summaries, including teens.
Last but not least, Google is experimenting with a dedicated AI search chatbot, akin to ChatGPT search mode and Perplexity. It's like Gemini but combines Google's real-time search capabilities for the most up-to-date responses. The new feature, AI Mode, is currently available in Google's testing ground called Labs. But it's an indication that Google Search might soon have only AI-generated search results.
Recommendations to put glue on pizza and eat rocks be damned, Google has signaled that injecting AI into all of its apps and services is the driving force of its business strategy. In the announcement, Google's VP of Search Robby Stein said, "People are using Google Search more than ever as they get help with new and more complex questions." But that obscures the fact that AI Overviews can't be turned off and it doesn't address the hallucinations that still plague the model and might never go away.
Former Mashable editor Mike Pearl did an audit on the first six months of AI Overviews and found that while it's fine for simple queries, it still hallucinates on more "uncommon queries" by misinterpreting what's found on the web. It also erroneously builds on faulty queries like using baking soda to thicken soup (which you definitely shouldn't do.)
"If the basis for a search is wrong or flawed, and the AI Overview doesn't catch the problem, then it stands to reason the user won't notice it either," said Pearl. That's to say, at best, it could weaken Google's reliability as a search engine, and at worst, it could reinforce misinformation.
Despite persistent inaccuracies that have become something of a running joke (seriously, just Google "Google AI search fails"), the company is barreling ahead with new experiments.
AI Mode is clearly an effort to compete with AI-powered search engines like Perplexity and ChatGPT search mode. These tools have capitalized on users leaning more into chatbots as a source of information, which threatens Google's core product.
In Labs, screenshots display AI Mode as a tab at the top of the Google Search app next to prominent filters like All, Places, Maps, and Images. According to the description page, the underlying model is Gemini 2.0, which has reasoning capabilities, meaning it breaks down queries into step-by-step instructions to search and verify.
This approach purportedly reduces hallucinations since it allows the model to check its work rather than spewing out the probabilistic next word.
Google One AI Premium subscribers who pay $20 a month get access first, and those enrolled in Labs can sign up for the waitlist.
