OPINION: Stop agreeing with me, Mr. AI

On May 19, the New York Times published an article with the incredible headline, “Book on Truth in the Age of A.I. Contains Quotes Made Up by A.I.” The article says the Steven Rosenbaum book “The Future of Truth: How AI Reshapes Reality” contains misattributed quotes which misrepresent speakers’ views on subjects.

The Times contacted Rosenbaum trying to figure out what was going on, and Rosenbaum confirmed. He said, “As I disclosed in the book’s acknowledgments, I used AI tools ChatGPT and Claude during the research, writing and editing process. That does not excuse these errors, of which I take full responsibility. I am now working with the editors to thoroughly review and quickly correct any affected passages; any future editions will be corrected.”

But, Rosenbaum also said the errors “[serve] as a warning about the risks of AI-assisted research and verification, that is why I wrote the book. These AI errors do not, in fact, diminish the larger questions that the book raises about truth, trust and AI and its impact on society, democracy and editorial.”

This is a wildly annoying way to defend your work, especially due to the flippancy a career writer would use a tool (arguably misuse it) and subsequently criticize it without a sense of self-awareness. The idea an editorial error is an AI’s fault, and not the individual who doesn’t do the due diligence of fact-checking, shows such an absurd sense of faith in an incredibly new institution.

AI hallucinations, the amazing term our society has come up with to describe how AI tries to make up facts in their informational blind spots, are sometimes amazingly funny.

A friend of mine was looking up nutrition facts for a fast food burger. When looking up the information, Google created an AI summary which included a fully confident fact that the burger was only 43 calories. This is a burger that contained two 1/3 pound beef patties, two slices of cheese, four slices of bacon and two buns. There’s probably 43 calories in one bite.

For anyone who has had a “conversation” with an AI, it’s always fun to push the limits of the tool. Saying “no, what you said was not true,” will almost always lead to the AI fiercely apologizing, promising it will better, then creates some more hallucinated facts.

Similar to a child with seemingly infinite energy but has learned how to lie, AI will try to downplay its mistakes and will always appeal to the good side of the user. It’s overly kind and toothless most of the time.

It’s part of the reason why so many users are now using AI chatbots for mental health-related conversations. Having a faceless being, who won’t tell anyone the secrets you send into the cloud (except maybe the megacorporation funding it) seems much easier than paying for therapy.

But, having an AI that will always agree with you is useless. I’m already a narcissistic person. If I needed someone to agree with me, I just use my imagination to place myself on a talk show and expound eloquently on whatever I want.

Part of the reason why people pay for therapy is because a therapist is trained to challenge their patient. Instead of agreeing on whatever input their patient gives them, they change perspectives and give feedback on how to process certain things. They can feel what their patient feels, and that’s essential to the connection therapy creates.

An AI who will constantly say, “it’s alright to feel this way” is useless and will have an effect of an ouroboros eating its own tail. And, so does using AI for research.

The positive reinforcement AI gets in the research process isn’t based on the information it digs up; it’s based on how the user will use the information dug up. Asking an AI to generate more detail on something when none might exist won’t lead to it explaining its own limits; it will just try to look as plausibly realistic as it can be to satisfy more positive reinforcement.

If AI was a dog fetching a stick, it will get a few pets for bringing back the stray twigs. But if no more trees are left, it will start eating the fence and the house, whatever looks believably like a stick. If one is blind, or doesn’t do the due diligence to make sure this dog isn’t destroying the foundation of truth, then they’ll never see when the house collapses.