Google has shared insights into what led to some strange and incorrect responses seen on its new AI Overviews feature last week. In a blog post, Liz Reid, VP of Google Search explained that data gaps and unrealistic queries tripped up the AI systems at times.
The tech giant rolled out AI Overviews with the goal of improving search experience for users. However, within days several users reported seeing bizarre summaries that did not make sense. Google acknowledges issues came up for uncommon questions where little verified information exists online.
Reid mentioned the AI struggled with nonsensical queries like “how many rocks should I eat”. Since no one actually searches for rock eating advice, the systems picked up irrelevant or satirical web content instead. Other unusual topics that stumped the AI included using glue to put cheese on pizza.
Misinterpreting language on websites and forums also led to some odd results. A spike in deliberately fake screenshots on harmful topics like leaving pets in hot cars further contributed to confusion.
In response, Google developed over a dozen technical changes after closely analyzing errors. Mechanisms have been added to detect silly queries and restrict using unreliable sources. Auto-generated content is now better screened for sensitive news and health advice.
While most user feedback has been positive, Google remains committed to closely monitoring all responses. The company will continue refining how and when AI Overviews display based on ongoing user reports. The recent launch issues highlight work still needed in interpreting obscure web content to ensure helpful, accurate information for people.
