Grok, the Artificial Intelligence used on Elon Musk's X, formerly Twitter, has been caught "lying" several times about pictures of malnourished children in Gaza. That's according to a report from fact-checkers RTVE, pointing to at least three times where users asked Grok about the veracity of pictures of starving children in Gaza, with very visible effects of malnutrition and illness, and X's AI said, wrongly, that the photos were not from Gaza... when they actually were.
"@grok, what year are these images from?", asked Kristy4TRUMP in response to a post from US Senator Bernie Sanders, last July 28. "These images are from 2016, showing malnourished children in a hospital in Houdieda, Yemen, amid the civil war there. They do not depict current events in Gaza", said Grok.
<social>https://x.com/Shayan86/status/1950590940764393815</social>
Except that they weren't: these are photos taken in Gaza on July 23, verified by Associated Press, as shared by Shayan Sardarizadeh, BBC reporter specialised in AI and disinformation.
On other occasion, Grok says that the picture of a mother carrying a baby with a plastic bag as a nappy and nothing but bone under his skin is from Yemen's 2017 famine, when in reality it's a picture taken by Ahmed al-Arini on July 21, with BBC even interviewing his mother, Hedaya al-Muta.
<social>https://x.com/grok/status/1948374632937185397</social>
Repeatedly, fact checkers have been finding that Grok is lying about the veracity of pictures, even when categorically claiming to be true. This shows the danger of using the AI as a verification tool, which can cause even more misinformation. Some experts believe that in Grok's case it is caused due to the AI using prompts trained on X, a social network where fake news and misinformation run free.
Experts asked by RTVE advise that, if we aim to use AI to combat misinformation, we must use more than one AI model and not limit to Grok, one that was intentionally made to have "unfiltered answers" and trained "not to be politically correct", which has, in many occasions, proved controversy for leaning to Musk's own political views, fuelling conspiracy theories including "white genocide" and using an "edgy" and "witty" tone.