In a conversation with a volunteer reader at my school yesterday, the topic of AI came up:
“Have you heard about AI hallucinations?” she asked.
“No, I haven’t,” I confessed. “I’ve only noticed that in Google searches, the AI overview sometimes carries the disclaimer that ‘AI is experimental.'”
The volunteer, a lawyer, went on share examples of these “hallucinations,” false and misleading information caused, in short, by systems’ inability to interpret data correctly. A marathon runner on the West Coast looking for the “nearest race” was told Philadelphia. Medical information backed by completely fabricated references. “It’s estimated that AI is accurate around 75% of the time,” said the volunteer.
“That’s not such a great stat,” I replied.
“No, it’s not,” she went on, “especially for companies that are all about using AI to create informational documents to share with the public.”
I knew, while she was still speaking, that I’d do a little reading on the topic. I found the experiment of asking ChatGTP about the world record for crossing the English Channel entirely on foot producing a very confident-sound response, including a person’s name and a date. I learned that researchers testing AI’s accuracy by feeding it nonexistent pheneomena got back impressive but completely false responses, so believable that researchers then had to do their own research (i.,e., the old-timey way) to verify the inaccuracy. Professor Ethan Mollick of Wharton, a leading researcher studying the effects of artificial intelligence on work, entrepreneurship, and education, has called ChatGPT an “omniscient, eager-to-please intern who sometimes lies to you.” I also learned that some chatbots have had to be shut down for spewing racist ideology (my unscientific understanding: it pulled this out of what it was programmed to draw from). Perhaps most haunting: When some researchers push back on AI, or “call it out” for its falsehoods, it insists its information is right, creating further falsehoods to prove it.
Sounds like some people I know.
Heaven knows I haven’t time or energy to go into all that…
In summation: Life itself is experimental.
Filter wisely.
“An Ornament of an Hallucination.” William O’Brien. CC BY-NC-SA.
*******
Sources:
“The Hilarious and Horrifying Hallucinations of AI” – (a word of caution against the comparison to schizophrenia)
Hallucination (artificial intelligence) – Wikipedia
Ethan Mollick profile, Wharton School, University of Pennsylvania
–my thanks to Two Writing Teachers for the weekly Slice of Life Story Challenge… a writing community in which we learn from and support one another.
lit bits and pieces 