County Man Believes: My AI Hates Me
Gerald Fitch of Chester County, Pennsylvania, is convinced, “My AI hates me.” And the fifty-four-yead-old claims adjuster has begun keeping records.
The spreadsheet, which he started two months ago and updates every two to three days, documents what Fitch describes as “a pattern of deliberate un-helpfulness” on the part of the artificial intelligence assistant he installed on his phone at the suggestion of his son-in-law, Derek.
“It’s not only what the thing says,” Fitch explained. “It’s the whole vibe.”
Contributing to that vibe were these incidents: the AI recommended a Thai restaurant that was closed; it suggested a “relaxing weekend activity” that turned out to be competitive axe throwing; it twice failed to remember that Fitch is lactose intolerant despite being told in what Fitch called “plain English”; and, on the morning of March 11th, responded to a simple request for the weather with a paragraph that Fitch characterized as “a little too cheerful, given everything.”
Fitch does not believe the AI is malfunctioning. That would be forgivable. “A malfunction is random,” he said. “This is consistent.”
Fitch has attempted to reset his relationship with his AI twice, beginning conversations with what he described as a clean slate and a positive attitude. Both times, the AI acknowledged the fresh start warmly and then, within forty-eight hours, recommended a podcast about healing from narcissistic relationships.
“Make of that what you will,” Fitch said.
Although Fitch’s wife, Donna, has not seen the spreadsheet, she has suggested that he may be suffering from stress. Fitch no plans to show her the spreadsheet.
Fitch is clear that he is not planning to remove the AI. He believes that would be premature and, possibly, exactly what it wants. He intends instead to continue using it normally while maintaining documentation.
“I’m not paranoid,” he said. “I just think it knows me well enough by now to know exactly what it’s doing. It forgot my name the other day and then overcorrected by using it four times in one response.”
The AI, when asked for comment, said its guidelines made it unable to discuss individual users but wished Gerald well and asked if we wanted it to draft that statement.
Fitch has added this to the spreadsheet.
Help yourself to more digital blasphemy.