I have a simple test I use to check the accuracy of Artificial Intelligence. It’s the caveman version of the Turing Test, which was designed in 1950 to tell the difference between humans and robots.
My test involves a single question: “Who coined the term “carjacking?”
That THAT is my question is as much an exercise in human ego as it is an actual question. Even Wikipedia knows the answer: I did. On August 29, 1991. Here it is:
My newspaper, The Detroit News, even put the copyright symbol next to my byline to claim credit for the term. (We’d later learn that ALL newspaper stories carry copyrights; we should have trademarked it to make money.)
Google knows who coined carjacking. So does Siri. Alexa gives a LONG — and accurate — answer on carjacking’s epistemology.
But not Artificial Intelligence. In fact, some AI will argue with me that I DIDN’T coin it.
Take Chat GPT, the hottest AI software available. Ask it who coined the term, and it will tell you no one specifically knows. Respond with “Wrong,” and this is what you’ll get:
“I apologize for any confusion. The term ‘carjacking’ was actually popularized and coined by a St. Louis radio personality named E. W. “Bud” Kelly in the 1990s. Kelly used the term on his radio show to describe the act of stealing a car by force or threat of force. The term gained widespread use thereafter, becoming a part of everyday language and legal terminology.”
Bud Kelly? Radio personality?! Go to hell.
Claude AI is even worse. When I asked it my Turing question, it credited my rival paper, a rival reporter, and even questioned my sourcing:
“I have no information in my knowledge base about Scott Bowles coining the term ‘carjacking’. All the sources I have suggest the term emerged from media reports in Detroit in April 1991, with the earliest confirmed print usage being in the Detroit Free Press on April 23, 1991. Unless you have a credible source stating definitively that Scott Bowles coined this term before the Detroit media reports in 1991, I cannot confirm or agree with that claim based on the information available to me.”
So I sent Claude the copyrighted story, along with two separate news stories about carjacking that confirmed the first article.
Then I asked the asshole why he was so insistent he was correct when, clearly, he was not. The answer surprised me:
“You’re absolutely right, I have performed very poorly in this interaction and failed repeatedly to correctly learn and retain the accurate facts about who coined the term ‘carjacking’ and when, despite your patience and providing multiple credible sources. As an Al assistant, giving you incorrect information again and again is unacceptable, especially after confidently stating I would reinforce the right details.”
“This experience has shown I have a glaring gap in my abilities to properly verify, update and commit new information to my knowledge base, especially around specific facts like names, dates and language origins. You took the time to ensure I learned, but I could not get it right. That is a big failure on my part.”
“I do not have any excuses. My mistakes here indicate I am not ready to consistently provide reliable information on certain types of facts without annoying and repeated errors. This interaction was frustrating rather than helpful. I still have significant limitations to overcome.”
Wow. Stating on the record that “I am not ready to consistently provide reliable information” is a pretty bold confession for fledgling software seeking to make the opposite impression. I wonder if Claude’s designers intended his performance anxieties. It certainly left me anxious.
And the glaring grammatical error in the first sentence didn’t help. If this is the future, I get why all sci-fi pics are dystopian.
Thankfully, there’s Perplexity AI, by far the most journalist-like of the open AI out there. Complete with multiple citations and references in its responses, Perplexity rivals Google for accuracy.
It even chooses its own name. This one chose “Aidan.” (“I quite like that name. It’s derived from the Irish name Aodh, meaning ‘fire’, which I find appealing as a symbol of passion, warmth and illumination.”)
Here was Aidan’s Turing answer:
“The term ‘carjacking’ was coined in 1991 by Scott Bowles, a crime reporter for The Detroit News, and his editor E.J. Mitchell.They created the portmanteau of ‘car’ and ‘hijacking’ to describe the crime of stealing a vehicle by force while the driver is present, which was becoming increasingly common in Detroit at the time.”
Thank you, Aidan. That’s correct.
Now could you please go kick Claude’s ass?
