The days of whackadoodle hallucinations from generative artificial intelligence might be few and far between now, but AI still can get things so completely wrong it borders on the comical.
Google is learning this the hard way as it rolls out AI-generated answers into search—and the internet is not letting things slide. Social media has been loaded with examples of the AI’s flubs, which range from the sublime to the absolutely ridiculous.
Leading the pack is the now-widely-shared example of Google suggesting you “mix about 1/8 cup of non-toxic glue into the sauce” to keep cheese from sliding off of your pizza slice. The source of this advice (which we strongly suggest you do not follow) is seemingly a joke post from Reddit made 11 years ago by a user whose name we can’t repeat in a family financial publication. (The answers are almost word for word.)
Ask Google which presidents went to the University of Wisconsin-Madison, and you’ll get an answer that not only makes history teachers weep, but could bring to mind a George Romero zombie movie. Andrew Jackson, for instance, who died in 1845, graduated in 2005, Google’s AI says. William Harrison, meanwhile, graduated in 1953 and 1974, which is especially impressive seeing as he died 31 days after his inauguration in 1841. Harry Truman, John Kennedy and others also graduated long after their death.
The source for this misinformation came from the school, actually—only Google AI failed to read the part of the page that said these were the graduation years of alumni with presidential names. No actual president has ever graduated from the school.
A Google spokesperson downplayed the examples going viral, saying, “The examples we’ve seen are generally very uncommon queries, and aren’t representative of most people’s experiences. The vast majority of AI Overviews provide high quality information, with links to dig deeper on the web. Our systems aim to automatically prevent policy-violating content from appearing in AI Overviews. If policy-violating content does appear, we will take action as appropriate. We conducted extensive testing before launching this new experience, and will use these isolated examples as we continue to refine our systems overall.”
Other examples of the AI’s goofs include:
Google's AI Overviews appear to be presenting comedic / sarcastic / satire content as fact pic.twitter.com/gsXf8Y8eyi
— Joe Youngblood (@YoungbloodJoe) May 21, 2024
Good ol’ Google AI: telling you to do the exact things you *are not supposed to do* when bitten by a rattlesnake.
— ern. (@ErinEARoss) May 19, 2024
From mushrooms to snakebites, AI content is genuinely dangerous. Pic.twitter.com/UZXgBjsre9
google’s AI is so smart, it’s honestly scary pic.twitter.com/7aa68Vo3dw
— Joe Kwaczala (@joekjoek) May 19, 2024
typing "1000km to [word]" is a surefire way to get the awful google AI to say something incredibly funny pic.twitter.com/eOOeb6d08E
— zach silberberg (@zachsilberberg) May 21, 2024