Liz Reid, the Head of Google Search, has acknowledged that the company’s search engine has produced some “odd, inaccurate or unhelpful AI Overviews” after they rolled out to everyone in the US. The executive provided an explanation for Google’s more peculiar AI-generated responses in a blog post, where the company also announced the implementation of safeguards to help the new feature return more accurate and less meme-worthy results.
Reid defended Google and noted that some of the more egregious AI Overview responses circulating, such as claims that it’s safe to leave dogs in cars, are fake. The viral screenshot showing the answer to “How many rocks should I eat?” is real, but she explained that Google produced an answer because a website had published satirical content on the topic. “Prior to these screenshots going viral, practically no one asked Google that question,” she explained, indicating that the company’s AI linked to that website.
The Google VP also confirmed that AI Overview suggested using glue to get cheese to stick to pizza based on content from a forum. She mentioned that forums usually provide “authentic, first-hand information,” but they can also lead to “less-than-helpful advice.” While she didn’t mention the other circulating viral AI Overview answers, as The Washington Post reports, the technology also told users that Barack Obama was Muslim and that people should drink plenty of urine to help pass a kidney stone.
Reid stated that the company extensively tested the feature before launch, but noted that “there’s nothing quite like having millions of people using the feature with many novel searches.” Google was able to identify patterns wherein its AI technology faltered by reviewing examples of its responses over the past few weeks. Consequently, it has implemented protections based on its observations, starting by enhancing its AI’s ability to better detect humor and satire. Google has also updated its systems to limit the integration of user-generated replies in Overviews, such as those from social media and forum posts, which could provide misleading or harmful advice. Additionally, it has “added triggering restrictions for queries where AI Overviews were not proving to be as helpful” and has stopped showing AI-generated replies for certain health topics.
