Jeff Chiu/AP
Alphabet CEO Sundar Pichai speaks at the Google I/O event in Mountain View, California, Tuesday, May 14, 2024. The company announced a new AI search results summary feature, but is now facing criticism over some incorrect results.
new york
CNN
—
Google promised that its new artificial intelligence search tool will “work for you” to make finding information online faster and easier, but just days after its release, the company has already walked back some of the inaccurate results.
Earlier this month, Google introduced an AI-generated summary tool for search results. The feature appears in search results to help users avoid clicking multiple links to get a quick answer to their question, but it has come under fire this week for providing false or misleading information to some users' questions.
For example, according to Google’s AI summary, president It's a common misconception that Barack Obama is a Muslim. In fact, he is a Christian. Another user wrote: The Google AI summary states that “of 54 African countries, none begins with the letter 'K',” apparently forgetting Kenya.
Google confirmed to CNN on Friday that the AI-generated summaries for both searches had been removed for violating the company's policies.
“The majority of our AI Overviews provide high-quality information with links to dig deeper on the web,” Google spokesperson Colette Garcia said in a statement, adding that several other viral examples of Google AI failings appear to have involved image manipulation. “We conducted extensive testing before launching this new experience, and as with other features we launch in Search, we appreciate your feedback and will take prompt action where appropriate, in accordance with our content policies.”
At the bottom of Google's AI search overview, it says “Generative AI is experimental,” and the company says it runs tests designed to mimic potential bad actors to ensure that false or low-quality results don't appear in its AI overview.
Google's search overview is part of a larger effort by the company to incorporate Gemini AI technology across all of its products as it tries to keep up in the AI arms race with rivals like OpenAI and Meta. But this week's blunder shows that adding an AI prone to confidently making false statements risks damaging Google's reputation as a trusted source for online information search.
Even for less serious searches, Google's AI summaries appear to sometimes provide incorrect or confusing information.
In one test, CNN asked Google, “How much sodium is in pickle juice?” The AI summary responded that 8 fluid ounces of pickle juice contains 342 milligrams of sodium, while less than half that amount (3 fluid ounces) contains more than twice as much sodium (690 milligrams). (Best Made pickle juice, sold at Walmart, contains 250 milligrams of sodium in just 1 ounce.)
CNN also searched for “data used for Google AI training.” In its response, the AI brief acknowledged that it was “unclear whether Google prevents copyrighted material from being included” in online data collected to train its AI models, noting larger concerns about how the AI company operates.
This isn't the first time Google has had to pull back features from its AI tools due to an embarrassing blunder: In February, the company suspended its AI photo generation feature that creates portraits of people after it was criticized for generating historically inaccurate images and overrepresenting people of color instead of white people.
Google's Search Labs webpage allows users in regions where AI Search Overview is available to turn the feature on or off.
