Most people who dispute an ethics investigation hire a lawyer or write an op-ed. Dale Caldwell, New Jersey's former lieutenant governor, went with chatbots. In a televised interview on NJ PBS, Caldwell said he ran the investigation report through "multiple AI platforms" — and that the machines cleared him.
The background: Caldwell was forced to resign on September 25 after an independent 61-page investigation, conducted by a respected former state attorney general, found he had sexually harassed a staffer and repeatedly violated state ethics rules. According to CBS New York's reporting on the probe, the findings included pushing for the hiring of a girlfriend without disclosing the relationship, taking personal guests to ticketed events on taxpayer money, and making a sexual comment to a staffer. Governor Mikie Sherrill called for his resignation, and her office said the investigation "substantiated serious violations of state ethics rules and found that his conduct constituted sexual harassment."
Caldwell has been making the media rounds to contest all of it. On NJ PBS with host Rob Nelson, he offered his most novel defense yet: "I put it through AI," he said, "and it said 59 times, it [sic] said, 'What would your findings be?' There was no instance in any of the AIs that said there would be a finding of sexual harassment."
As evidence goes, this is less compelling than it sounds. Chatbots are notoriously sycophantic — trained to be agreeable, they routinely tell users what they want to hear, and the output depends heavily on how the question is framed and which parts of a document are included. Running a document through chatbots and asking "what would your findings be" is not an audit; it is a procedure that can be repeated, adjusted and re-run until the answer feels right. As The Verge put it bluntly, it seems highly unlikely that Caldwell would want to hear an AI confirm findings against him.
The deeper problem is the aura. When a public official says "I AIed it," the phrase borrows the credibility of computation for what is, at bottom, a pattern-matching tool with no access to evidence, testimony or legal standards. A machine cannot conduct an investigation, weigh witness credibility, or apply a legal definition of harassment. Presenting its output as a verdict — 59 times or 59,000 times — launders a partisan claim through what sounds like a neutral process.
Caldwell says he is exploring legal remedies, including a challenge through the New Jersey Civil Service Commission, to clear the harassment findings. He will have to do so with human adjudicators, under rules of evidence that do not accept "the chatbot agreed with me" as proof. But the episode is a preview of a defense strategy that courts, employers and ethics boards are about to see much more often — and it arrived, fittingly, on live television.
Compiled from The Verge and CBS News New York.
Comments (0)
Log in to join the discussion
Log InNo comments yet