Google’s AI Overview Missteps: From Rock Eating Recommendations to Gluey Pizzas and Beyond

To go back to this article, go to My Profile, and then click on View saved stories.

Reece Rogers

Google Acknowledges Failures in Its AI Summaries Search Feature

In response to the widespread sharing of peculiar and inaccurate responses produced by Google’s latest AI Summaries feature on social media, the company initially minimized concerns over the technology’s shortcomings. However, by late Thursday, Google’s search leader, Liz Reid, conceded that these mishaps had underscored the need for enhancements, stating, “We wanted to clarify what transpired and the measures we’ve implemented.” Reid’s message specifically pointed out two notably erroneous and popular AI Summary outcomes. These included Google's system mistakenly advocating for the consumption of rocks by claiming it “can be beneficial to your health,” along with recommending the use of harmless glue as a method to thicken pizza sauce.

The subject of consuming rocks hasn't been widely discussed or queried on the internet, resulting in a scarcity of references for search engines to utilize. Reid mentioned that the AI system identified a piece from The Onion, a website known for its satire, which a software firm had shared, and mistakenly regarded the content as true.

Regarding the incident where Google advised its users to use glue on pizza, Reid explained the mistake as a breakdown in detecting humor. She mentioned, "We encountered AI Summaries that included content from discussion forums which was sarcastic or meant as a joke," she noted. "While forums can be a valuable resource for genuine, direct information, they sometimes offer advice that's not very practical, such as applying glue to ensure cheese adheres to pizza."

It's advisable to thoroughly review any dinner menu created by AI before finalizing it.

Reid recommended not to assess Google's refreshed search approach merely from popular screenshots, stating it would not be just. She mentioned that the company carried out thorough testing prior to its rollout, and according to their findings, users appreciate the AI Summaries, evidenced by an increased likelihood of users remaining on a page they found through this method.

What explains the mortifying mishaps? Reid described the errors that garnered attention as stemming from a comprehensive online review that didn't always have good intentions. "Experiencing millions of users engaging with the feature through a multitude of unique searches is unparalleled. We've also encountered bizarre new queries that appear designed to generate incorrect outcomes."

Authored by Tess

Authored by Will

Authored by Kim Zetter

Authored by Hannah Singleton

Google has stated that certain screenshots circulating widely, showing AI Overviews responding incorrectly, are not authentic, a claim that appears to be accurate following tests conducted by WIRED. For instance, a screenshot shared on X suggested that an AI Overview had affirmatively answered the bizarre question, "Can a cockroach live in your penis?" indicating that such a scenario is common. This screenshot has attracted over 5 million views. However, upon closer examination, it became clear that the format of the screenshot did not match the actual presentation of AI Overviews. Furthermore, WIRED's attempts to replicate the result were unsuccessful.

It wasn't only individuals on social media platforms who were deceived by the false images purporting to represent AI-generated summaries. The New York Times had to amend its coverage regarding the tool, explaining that the AI Summaries did not actually advise individuals to leap off the Golden Gate Bridge if they were feeling depressed—this was merely a grim joke circulating on social media. “It has been suggested by some that our search results included harmful advice on subjects such as locking dogs in vehicles, smoking during pregnancy, and dealing with depression,” Reid mentioned in a statement last Thursday. “However, such AI-generated summaries were never produced.”

However, Reid's message also highlights that the initial version of Google's significant search enhancement was not without its issues. She mentioned that the company implemented "over a dozen technical adjustments" to AI Overviews.

Four key improvements have been highlighted: Enhancing the ability to identify and disregard irrelevant queries that do not merit an AI-generated summary; reducing the dependence on content sourced from user-driven platforms such as Reddit for generating these summaries; limiting the frequency of AI Overviews in contexts where they have not proven beneficial to users; and bolstering the safeguards that prevent the generation of AI-driven summaries for critical subjects like health matters.

Reid's blog post did not include any discussion about major reductions in the AI summaries. Google has stated its commitment to keep an eye on user feedback and make necessary adjustments to the features accordingly.

Recommended for You…

Direct to your email: Dive into the future with Will Knight's Fast Forward, focusing on AI breakthroughs.

Indian electorates are being overwhelmed by countless deepfake videos

Incarcerated individuals purchased tablets behind bars, only to discover an unful

The single factor limiting the progress of heat pumps

Eternal Sunshine: Discover the Top Sunglasses for Any Expedition

Lauren Goode

Steven Levy

Knight Will

Paresh Dave

Knight Will

Lauren Goode

Reported by:

Knight, Will

Additional Content from WIRED

Critiques and Manuals

© 2024 Condé Nast. All rights reserved. When you buy products through our links, WIRED might get a share of the sale via our affiliate relationships with stores. You’re not allowed to copy, share, send, store, or use the content from this website in any way without getting explicit written approval from Condé Nast. Choices about Ads.

Choose a global website