Google breaks its silence on AI search results that told people to eat rocks and put glue on pizza
While Apple’s WWDC2024 is expected to affect many AI-focused featuresLast week, Google gave us some insight into what we shouldn’t do.
The company, whose search engine is regularly algorithmically updated to add new features to adjust rankings, introduced its AI Overviews feature to summarize answers to a user’s query.
However, because the AI was seemingly unable to distinguish between genuine medical advice and satire, the process went more than a little awry, for example suggesting that users attach cheese to their pizza with glue or eat rocks to maintain a healthy lifestyle.
In a blog post this week According to Liz Reid, vice president and head of Google Search, the company had a challenging start.
Google discusses introduction of AI Overviews
While the post suggests there are some “fake screenshots,” Reid notes that AI Overviews mostly fell for “queries that people don’t normally ask.”
“One area where we identified was our ability to interpret nonsensical questions and satirical content,” it says, referring to the question “How many rocks should I eat” that went viral.
“There is also not much web content that seriously addresses this question. This is often referred to as a “data gap” or “Information gap”, where there is a limited amount of high-quality content on a topic,” explains Reid.
“In this case, however, there is satirical content on the topic … which also happened to be republished on the website of a geological software vendor. So when someone typed that question into search, an AI summary appeared that faithfully linked to one of the few websites that addressed the question.”
Google promises that improvements have been made
According to the post, Google has now updated its technology and “introduced better detection mechanisms for nonsense queries that should not be shown an AI digest, and limited the inclusion of satirical and humorous content.” At the same time, it has moved away from user-generated content that could be considered misleading.
“We’ve added trigger restrictions for queries where AI summaries have not proven to be as helpful,” Reid adds, also pointing to “additional trigger refinements to improve our quality protections” when it comes to health and news-related content.
It is clear that Google’s AI overviews need a significant change to bring them up to date – otherwise it could Join the Google Product GraveyardThere are also concerns that by serving content without encouraging clicks to websites, Google will ultimately cannibalize the websites visited by users.