Google last week unveiled new artificial intelligence features to answer people's questions and announced its biggest search changes in years, as part of the company's attempt to catch up with rivals Microsoft and OpenAI.
The new technology has since spawned a number of false claims and inaccuracies, from recommending the glue as part of a pizza recipe to consuming the rocks for nutrition, damaging Google's reputation and causing uproar online.
The incorrect answers in a feature called “AI Overview” have undermined trust in the search engine that more than 2 billion people rely on for reliable information. The backlash comes as Google comes under greater pressure to safely embed AI in its search engine, as other AI chatbots have lied or behaved oddly.
The release also broadens a pattern of Google running into trouble shortly after releasing its latest AI features. In February 2023, Google shared misinformation about space when it launched Bard, a chatbot to rival ChatGPT. The company's market capitalization subsequently fell by $100 billion.
In February, the company released Bird's successor, Gemini, a chatbot that generates images and acts as a voice-controlled digital assistant. Users quickly noticed that the system almost always refused to generate images of white people and was inaccurate in its portrayals of historical figures.
After each of these scandals, tech industry insiders have criticized the company for failing, but in interviews, financial analysts said Google needs to act quickly to keep up with rivals, even if it means growing pains.
“Google doesn't have a choice right now,” Thomas Monteiro, a Google analyst at Investing.com, said in an interview. “Companies need to move very quickly, even if it means skipping a few steps along the way. The user experience needs to keep up.”
Google spokeswoman Lara Levin said in a statement that most AI Overview searches return “high-quality information, including links to explore further on the web.” AI-generated results from the tool typically appear at the top of the results page.
“Many of the examples we've seen are unusual queries, and we've also seen examples that have been doctored or that can't be reproduced,” she added. The company plans to use “isolated examples” of problematic answers to improve its system.
Google has been under pressure to integrate AI into popular apps since OpenAI released its ChatGPT chatbot in late 2022, which became an overnight sensation. But it's hard to control large-scale language models that aren't programmed like traditional software but instead learn from vast amounts of data pulled from the open web, including false and satirical posts.
(The New York Times sued OpenAI and its partner Microsoft in December, alleging copyright infringement of news content related to the AI system.)
Google unveiled its AI Overview with a bang at its annual developer conference, I/O, last week, and for the first time, the company is incorporating its latest large-scale language AI model, Gemini, into its flagship product, its search engine.
AI Overview combines statements generated from a language model with snippets from live links on the web, and while it can cite sources, it doesn't know if they're wrong.
The system was designed to answer more complex and specific questions than a typical search, which the company says will allow the public to benefit from all of Gemini's features while reducing the friction of searching for information.
However, things quickly went awry, with users posting screenshots of the problematic examples to social media platforms such as X.
The AI Overview instructed some users to mix a non-toxic adhesive into their pizza sauce to stop the cheese from sliding off, which appeared to be a fake recipe borrowed from an 11-year-old tongue-in-cheek Reddit post. The AI advised other users to take at least one stone a day for vitamins and minerals, advice that originated from a satirical post in The Onion.
The company's cash cow, Google Search, “is the only property Google needs to remain relevant, trustworthy, and useful,” Gergely Orosz, a software engineer who publishes the tech newsletter Pragmatic Engineer, wrote on X. “But examples of AI summaries turning Google Search into garbage are all over my timeline.”
Also shared was an example of Google instructing users in bold type to clean their washing machines with “chlorine bleach and white vinegar,” which can release harmful chlorine gas when mixed together, and in smaller type instructing users to clean with one first, then the other.
Social media users have been competing to see who can share the most outlandish response from Google, with some even doctoring the results. In one doctored screenshot, Google appears to quote a Reddit user as saying that a good cure for depression is to jump off the Golden Gate Bridge. Google spokesman Levin said the company's systems have never returned such results.
But AI Overview struggles with presidential history, noting that 17 presidents have been white and that Barack Obama was the first Muslim president, according to a screenshot posted to X.
It also states that Andrew Jackson graduated from college in 2005.
Kevin Roose Contributed report.
