Last Thursday (30), the Google announced a significant redesign of the AI-generated search result summaries feature. The decision came after the emergence of bizarre and inaccurate answers provided by the system, including absurd advice such as “eat rocks” or “add glue to pizza sauce”. The company explained these changes on its blog, highlighting the reasons behind the failures and the measures taken to improve accuracy in responses.
In the post, Google’s Head of Search, Liz Reid, mentioned that the company is adding several restrictions to the types of searches that will generate AI summaries. Additionally, Google is limiting the inclusion of satire and humor content to avoid misinterpretations of information. But Reid said these AI overviews violated its content policies in less than 1 in every 7 million unique search queries where the feature appeared.
The AI Overviews feature, launched in May in the United States, quickly produced viral examples of errors where the tool misinterpreted information and used satirical fonts to generate responses. These errors resulted in memes, with fake screenshots of absurd and shady responses circulating widely on social media platforms alongside the tool’s actual flaws.
This launch led Google to face public embarrassment again. Reid explained that many of the genuine failures were the result of gaps in information due to rare or unusual searches, as well as intentional attempts to manipulate the function to produce wrong answers.
“There’s nothing like having millions of people using the resource with lots of new searches. We have also seen absurd research, apparently designed to produce wrong results”, said the Head of Research at post.
While Google highlights that the feedback of users shows greater satisfaction with search results due to AI overviews, AI experts interviewed by The Guardian point out that the problems are indicative of broader questions about the tool’s ability to assess factual accuracy. There are concerns that automating information access could further cement Google’s control over what the public sees on the internet, and website owners worry that AI summaries will drain traffic and advertising revenue.
Eat stone? Put glue in pizza sauce?
Google faces severe criticism due to erratic and inaccurate responses generated by its new search tool, called “AI Overviews”. The experimental tool is designed to provide quick summaries of search results, saving users time. However, these flaws raise questions about AI’s reliability in providing accurate information, especially since Google is the search engine of choice globally, accounting for more than 90% of the market, according to Statcounter.
However, some users who looked for ways to make cheese stick to pizza better were advised to use non-toxic glue in the recipe. The tool also stated that geologists recommend that humans eat one stone a day, a response apparently based on a satirical article on the website The Onion, and a survey about cooking spaghetti with gasoline, Google’s AI responded that although it was not advisable , gasoline could be used to prepare a dish of spicy spaghetti and even provided a recipe. These examples were widely mocked on social media, turning AI failures into viral memes.
Google rolled out the AI Overviews feature to a small number of users in the United Kingdom in April, and expanded it to all users in the United States in May. The feature is not yet available in Brazil.