Google LLC’s new AI Overviews feature in Search has come in for some heavy criticism after generating a series of nonsensical, inaccurate and sometimes even potentially dangerous results in response to user’s queries.

When Google debuted the new feature at its Google I/O conference last week, it warned that it’s an experimental experience, based on generative artificial intelligence. Nevertheless, the dubious responses it has generated will likely be a concern for the company, given the threat that other AI companies pose to its dominant position in internet search.

Some of the responses it has come up with are downright laughable, with one user posting on X, formerly Twitter, that AI Overview’s helpful suggestion to prevent cheese sliding off a homemade pizza was to “mix glue with the cheese.”

Google is dead beyond comparison pic.twitter.com/EQIJhvPUoI

— PixelButts (@PixelButts) May 22, 2024

That particular answer appears to have been sourced from a 10-year-old Reddit post made by a user called “fucksmith,” and anyone who can comprehend the original comment will understand that it was clearly a joke.

But that unappetizing suggestion wasn’t the most harmful or dangerous recommendation it has made. In another example, AI Overviews told a user asking it how to sanitize a washing machine to create what is essentially deadly mustard gas, even providing instructions on how to do so.

I asked how to clean my washing machine, and it told me to make chlorine gas pic.twitter.com/BniM2j1VGv

— goose queef (@Mia62651937) May 15, 2024

Some of the more ridiculous results from AI Overviews include the claim that former U.S. President James Madison graduated an impressive 21 times from the University of Wisconsin, that Batman is actually a police officer, that snakes are mammals and that a dog has competed in major league sports such as the NFL, NHL and NBA.

There are concerns that AI Overviews is plagiarizing some of its responses too, with some appearing to be copied almost word for word from their original sources. Another user posted on X that Google simply added the term “my kid’s favorite” to a smoothie recipe that was taken from a website, but otherwise using exactly the same language.

The criticism of Google is perhaps somewhat harsh, given that just about every AI model made available to the public has been found guilty of making similar mistakes. But then again, Google is the first company to deploy generative AI at such an enormous scale, with the AI Overviews feature being made available to every user in the U.S. The frequency of the hallucinations suggests that the feature is far from being ready for prime time, despite Google’s claim that it represents the future of search.

The mistakes aren’t the only source of criticism, for others have expressed concern that it could be damaging to many smaller websites. Google said at I/O that the feature will benefit smaller sites as it will surface more of their content, as opposed to its traditional search engine, which tends to give more weight to better known websites. “We see that the links included in AI Overviews get more clicks than if the page had appeared as a traditional web listing for that query,” the company said at launch.

However, web users have a tendency to be lazy, and the worry is that by presenting the information people are looking for in an AI Overview summary, it will actually prevent those searchers from clicking through to the actual source of the content, starving them of traffic.

Those fears have prompted publishers to fight back. One organization, the News Media Alliance, which represents publishers, has called for increased action against AI models that are trained on web content without the permission of its owners.

“AI companies recognize the value of this content, evidenced by marketplace arrangements, and they rely on quality content to train their systems,” NMA President and Chief Executive Danielle Coffey said in a statement on legal action against another generative AI company, OpenAI. “The continued unconstrained use by AI companies of publishers’ valuable content without proper compensation is unlawful.”

Unfortunately, Google doesn’t seem to have made it possible for users to turn off the new feature. But those who are fed up with AI Overviews can sidestep its suggestions by going to Google Search, searching for any query, and then clicking on the “More” tab directly below the search box, followed by “Web,” which has the effect of filtering out the AI responses for that query.

An alternative solution is to install the Bye, Bye Google AI extension in Chrome, which automatically blocks any AI-generated responses to search queries. There are similar extensions available for the Firefox browser.

A more drastic option is to select a different default search engine, such as Microsoft Corp.’s Bing or the privacy-focused DuckDuckGo, but be warned that they aren’t always reliable, either.

Image: SiliconANGLE/Microsoft Designer