Last week, Google unveiled its biggest search shakeup in years, rolling out new artificial intelligence features to answer people's questions as it seeks to keep up with rivals Microsoft and OpenAI.
The new technology has since spawned a slew of falsehoods and errors, including recommending glue as part of a pizza recipe and eating rocks for nutrients, giving Google a black eye and sparking outrage online.
The feature's incorrect answers, called AI Overview, undermined trust in the search engine that more than 2 billion people turn to for authoritative information. And while other AI chatbots have been lying and behaving strangely, the backlash has shown that Google is under more pressure to safely integrate AI into its search engine.
The launch also extends a pattern of Google running into trouble immediately after rolling out its latest AI features. In February 2023, Google announced Bard, a chatbot to combat ChatGPT, sharing misinformation about the universe. Since then, the company's market value has fallen by $100 billion.
This February, the company launched Gemini, the successor to Bard. Gemini is a chatbot that can generate images and act as a voice-activated digital assistant. Users quickly realized that the system refused to create images of white people in most cases and drew inaccurate depictions of historical figures.
After each incident, tech industry insiders have accused the company of dropping the ball. But in interviews, financial analysts said Google must move quickly to catch up with rivals, even if it experiences growing pains.
“Google has no choice right now,” Thomas Monteiro, a Google analyst at Investing.com, said in an interview. “Companies have to move very quickly, even if it means skipping a few steps. User experience will have to catch up.”
Google spokeswoman Lara Levin said in a statement that the majority of AI Overview queries resulted in “high-quality information with links to dig deeper across the web.” AI-generated results from a tool are typically displayed at the top of the results page.
“Many of the cases we saw were uncommon queries, and we also saw cases that were manipulated or unreproducible,” he added. The company will use “isolated cases” to answer problems to improve its systems.
Since OpenAI launched its ChatGPT chatbot in late 2022 and became an overnight sensation, Google has been under pressure to integrate AI into its popular apps. But rather than programming it like traditional software, there are challenges in taming large-scale language models that learn from massive amounts of data pulled from the public 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 presented an AI overview last week at its annual developer conference, I/O. For the first time, the company has connected its latest large language AI model, Gemini, to its most important product: a search engine.
AI Overview combines statements generated from a language model with snippets pulled from live links from across the web. You can cite a source, but you won't know if the source is incorrect.
This system is designed to answer more complex and specific questions than general searches. As a result, the public can enjoy all the benefits Gemini can do and save some of the effort it takes to search for information, the company said.
But the situation quickly escalated, with users posting screenshots of problematic cases on social media platforms like X.
The AI brief instructed some users to mix non-toxic glue into their pizza sauce to prevent the cheese from slipping, a fake recipe that appeared to be borrowed from an 11-year-old Reddit post that was intended as a joke. AI instructed other users to take at least one stone a day for vitamins and minerals. This is advice that originated in a satirical post in The Onion.
Google Search, the company's cash cow, is “one asset that Google needs to remain relevant/trustworthy/useful,” Gergely Orosz, a software engineer in the tech newsletter Pragmatic Engineer, wrote about X. AI Overview is trashing Google searches all over my timeline.”
People also shared an example of Google telling users in bold font to clean their washing machines using “chlorine bleach and white vinegar.” When combined, these mixtures can produce harmful chlorine gas. In smaller fonts, users were told to use one and then the other.
Social media users have been trying to compete with each other to see who can share Google's craziest reactions. In some cases, the results were manipulated. One of the doctored screenshots appears to show Google quoting a Reddit user as saying that a good cure for depression is jumping off the Golden Gate Bridge. Mr. Levin, the Google spokesman, said the company's system never returned the results.
But the AI overview grappled with presidential history, saying 17 presidents have been white and Barack Obama was the first Muslim president, according to a screenshot posted on X.
Andrew Jackson also revealed that he graduated from college in 2005.
kevin rouge contributed to the report.