As widely recognized as the technology giant at the forefront of the battle against Microsoft/OpenAI, Google once again rose to the forefront of public opinion due to its new AI features.
At last week's annual I/O developer conference, Google announced in a high-profile manner the official introduction of AI functionality into its core search business, launching a feature called "AI Overviews", which will first be launched in the US market. In short, when users search, the first thing that appears in front of them is no longer a webpage link, but a piece organized by AI.
(Source: Google I/O)
However, as soon as this feature was launched, it "flipped over" - in addition to the common "hallucinations" of AI chatbots, a more serious problem is that Google's AI frequently popularizes "post bars" and satirical jokes on news websites as serious knowledge to users.
Pour some glue on the pizza!
As the most widely spread case on the Internet these days, when Google "AI Review" replied that "cheese cannot stick to pizza", it even suggested that "one eighth cup of non-toxic glue should be added to the filling". According to research by netizens, the source of this sentence is a rather sarcastic response from a netizen on Reddit, an American forum, eleven years ago.
(Source: X)
For example, when a netizen asked "how many stones should I eat every day," Google AI replied solemnly, "According to geologists' research, I should eat at least one small stone every day."
(Source: X)
The expression of this paragraph comes from The Onion, a website in the United States. The main business of this website is to rely on real background settings, fabricate a series of exaggerated and satirical articles to attract attention, and many events are completely fictional. So when senior netizens hear "Onion News", they know that the story is definitely not credible, but Google's AI has no discerning ability at all.
What makes American netizens even more vulnerable is that Google's AI can even confuse information about the President of the United States. For example, some netizens have stated that when asked how many Muslim presidents there are in the United States, Google will reply with "one, Barack Obama.". It is worrisome that this answer is not only untrue, but also likely derived from conspiracy theories spread by some far right groups on the internet.
In addition, Google's AI also suggested to netizens that "one can stare at the sun for 5-15 minutes, and people with darker skin can stare for up to 30 minutes", and happily admitted that "Google has violated US antitrust laws", claimed that "Batman is a police officer", and "a dog has been in the NBA." A series of absurd answers such as "appearing in NFL and other competitions".
Although it is not uncommon for AI chat robots to give wrong responses, the quality of the answers returned by default from the search engines commonly used by hundreds of millions of people is so outrageous that it quickly triggered public opinion on the Internet.
Google seems not ready to "withdraw"
According to the technology media The Verge, Google spokesman Meghann Farnsworth said in a statement that the errors mentioned on the Internet come from "very rare queries" and do not represent the experience of most people. The company has taken action to crack down on policy violations and is currently using these isolated cases to continue improving its products.
Google further stated that "AI Overview" can provide high-quality information in the vast majority of cases, accompanied by detailed webpage links. At the same time, the company said that some cases on the Internet could not be reproduced, and there might be deliberate tampering.
The accuracy of Google AI has been questioned by American society, partly due to a "warning from the past" - in February of this year, Google launched the Gemini "Wenshengtu" tool, but it was soon discovered that there was a serious problem of "fictional history". After the outbreak of public opinion, Google immediately announced the suspension of generating character images. Hassabis, the head of Google AI, stated in February that "the feature will be re launched within a few weeks," and there has been no further action so far.
In addition to a series of errors, Google's AI Review has also been pointed out to have other issues and risks.
For example, AI always excerpts content from different websites and makes slight modifications, which is also criticized as "large-scale plagiarism.".
Another issue is that since Google's AI is essentially "self searching and writing content," the company may have to bear legal responsibility for the false content generated by AI. Previously, according to Article 230, Google, as a search platform, was not responsible for the webpage searched.