Google is working to remove a wave of incorrect or even dangerous answers from its controversial AI-powered search results, including a response that suggested users eat rocks for nutrition.
The major tech company has faced backlash for spreading misinformation since the AI tool, known as AI Overviews, was launched in the US this month, expected to reach over one billion users by the end of the year.
In a notable example, AI Overviews allegedly answered the query, 'How many rocks should a child eat?' by inaccurately stating that UC Berkeley geologists recommend 'Eating at least one small rock per day.' The AI search results also prompted users to 'Put glue on their pizza' before consumption.
While, Google swiftly removed some inaccurate AI results, referred to as AI Overviews, the damage and meme-ification had already occurred.
Following this incident, in a company blog post published Thursday, Google’s head of search, Liz Reid, attributed the inaccurate AI results, known as AI Overviews, to 'data voids' and people asking odd questions. Reid asserted that AI Overviews are leading to 'higher satisfaction' with Search, arguing that the AI generally doesn’t 'hallucinate' but occasionally misinterprets existing web content.
'There’s nothing quite like having millions of people using the feature with many novel searches,' she wrote. She also noted the appearance of 'nonsensical new searches' seemingly intended to produce erroneous results and acknowledged the spread of numerous 'fake screenshots' of AI Overviews online.
The company's blog also explains how they are trying to fix AI Overviews by limiting when they appear for 'nonsensical' queries and satire. Part of Reid’s blog also compares AI Overviews to another longstanding search feature called featured snippets, which highlight information from a relevant webpage without using generative AI. According to Reid, the 'accuracy rate' for featured snippets is 'on par' with AI Overviews.