Last week, Google’s AI Overviews search feature faced a failure, providing inaccurate and comical answers to user queries. AI Overviews, an AI tool developed by Google, uses data from the internet to answer search requests without the need for users to hop through multiple websites.
Google recognised the issue and stated that an improvement is underway. Liz Reid, Head of Search, acknowledged in a blog post that while AI Overviews has been successful, there have been several glitches. She mentioned that the strange results were primarily due to nonsensical queries or manipulated screenshots. Reid insisted that many of the purported AI Overviews had never appeared, urging people to verify these instances by carrying out their own searches.
The primary reasons for the tool providing odd, inaccurate, or unhelpful results were the lack of quality data related to the queries or nonsense queries not detected by the search tool, according to Reid. The tool’s response to a bizarre question—”how many rocks should I eat?”—exemplifies its limitation. The answer was based on a satirical article from The Onion that humorously suggested eating a small rock daily. AI Overviews struggles to differentiate between humorous, satirical, and factual content.
Steps have been taken to improve the quality of results from AI Overviews. These include better detection mechanisms for nonsensical queries and satirical content, limitation of user-generated content utilisation in responses that could provide misleading advice, implementation of restrictions in queries where AI Overviews were deemed unhelpful, and the adding of robust guardrails to limit responses related to sensitive topics like news and health.
Delivering consistently accurate responses to search queries is a complex task. Despite its recent hiccups, AI Overviews is useful. However, it’s critical to approach the results with a healthy dose of skepticism. AI doesn’t create information; instead, it relies on human-generated information available on the internet. This brings forth an intriguing challenge both AI and human face when reading online content: discerning its veracity. Paradoxically, AI, intended to solve this issue, is making it more challenging.
As such, Google is working to fix the shortcomings in its AI Overviews search feature, acknowledging its recent failure. The company remains optimistic about the tool’s potential despite its current flaws.
