Google Pushing Its Unsafe Search AI on Users Who Didn’t Opt In

Has anyone spotted Bono at the Google HQ?

All Aboard

Don't want Google's experimental AI search feature embedded into your search pages? Too bad.

According to Search Engine Land, Google has started unleashing its AI-powered Search Generative Experience (SGE) product, which was previously available for users only on an opt-in basis, into the browsers of users who didn't choose to partake in Google's AI search experiment.

Google has defended the unsolicited SGE-ification of its platform, telling Search Engine Land that it's thus far only incorporated SGE automatically into a "subset of queries" that take up a "small percentage of search traffic in the US" and arguing further that the rollout will allow them to glean feedback from the users who, again, didn't elect to opt into the generative AI search service.

Regardless of how limited the effort is, though, the fact that Google's tentatively starting to test SGE in its open waters feels like a sign that the search giant's vision for an AI-infused gateway to the internet is still gunning full-steam ahead.

Complex Queries

For the uninitiated, SGE is a large language model-powered chatbot that sits at the top of Google results pages, where it gobbles web results and recapitulates them into organized, paraphrased answers for various queries. It'll generally link back to sources, though its sourcing is often flawed. Less-than-trustworthy user-generated content from sites like Reddit or Medium will sometimes be compiled into answers, for example.

Today, when we tested the query "today's news overview," SGE indeed provided a list of news bullet points, but linked out only to the homepages of various news sites to support its paraphrased roundup — as opposed to linking to the exact articles that might offer a Google user more useful information.

SGE has also been caught churning out plenty of incorrect information and deeply problematic takes, and was recently found to be sourcing information from — and linking back to! — to malware-laden spam. And broadly speaking, SGE even stands to change how our current internet and media landscapes function as a whole: if Google's AI is helpfully paraphrasing web-published content, where's the incentive for users to actually click the AI-feeding blue links?

In a statement to Futurism, a Google spokesperson noted that some of the AI errors in question, particularly the recent spam incidents, "only showed up for uncommon queries" and "in cases where an SGE user chose to manually generate an AI overview." They also added that Google utilizes its "core anti-spam protections to safeguard SGE from low-quality content for the vast majority of queries."

Per Search Engine Land, Google is first testing SGE in search queries where it believes an AI-paraphrased roundup might be particularly helpful or useful to users, including queries that are "more complex" or "involve questions where it may be helpful to get information from a range of web pages."

It could be argued that complex questions are ones that Google would do best to keep AI out of for now, but we digress. For now, don't be too surprised if you see an AI-generated search response crop up on your browser sometime in the near future.

Updated with a statement from Google.

More on Google SGE: Google's AI Search Caught Pushing Users to Download Malware