In a twist of irony that truly highlights the peculiar challenges of our digital age, Steven Rosenbaum, an author who meticulously crafted a book urging caution about artificial intelligence’s potential to warp truth, now finds himself entangled in the very web of deception he sought to unravel. His new book, “The Future of Truth,” was intended as a beacon, guiding readers through the labyrinthine ways AI is reshaping our understanding of reality, trust, and even the very fabric of public discourse. Yet, it appears that the siren song of AI, the very tools he warned against, may have lured him into a treacherous trap. The New York Times, in a recent report, revealed that Rosenbaum’s book, ironically, contains fabricated and misattributed quotes, seemingly conjured by the same chatbot tools he so painstakingly dissected within its pages. It’s a scenario almost too perfectly ironic to be true – an author sounding the alarm about AI-generated falsehoods, only to have his own work marred by them.
Rosenbaum, to his credit, has acknowledged the unsettling revelations, admitting that his book indeed contained “a handful of improperly attributed or synthetic quotes.” He’s initiated an internal investigation, a veritable soul-searching mission, to understand how these digital apparitions managed to slip into his final manuscript. In a statement, he disclosed, “As I disclosed in the book’s acknowledgments, I used A.I. tools ChatGPT and Claude during the research, writing and editing process.” While he accepted full responsibility for these glaring errors, he couldn’t help but acknowledge the undeniable irony. It’s as if the very subject he was studying reached out from the digital ether and left its mark, a mischievous prankster’s signature on his earnest endeavor. He’s promised that future editions will undergo a rigorous review and correction process, a commitment that speaks to his desire to uphold the very standards of truth he champions.
The impact of these “AI errors” is perhaps best illustrated by the bewildered reactions of those falsely quoted. Consider Kara Swisher, a prominent tech journalist, whose name was attached to a quote in a chapter detailing the perils of AI-generated falsehoods. The book claimed she said, “The most sophisticated A.I. language model is like a mirror. It reflects our own morality back at us, polished and articulate, but ultimately empty behind the surface.” Swisher, upon reading this, was understandably perplexed, stating unequivocally, “Never said that.” She even poked fun at the fabricated quote, quipping that it made her sound “unbearably pretentious” and like she had “a stick up my butt, according to ChatGPT.” Her reaction underscores the deeply personal and often embarrassing nature of such misattributions, where an author’s voice is stolen and twisted into something unrecognizable. It’s not merely a factual error, but an affront to intellectual integrity and personal brand.
Another stark example comes from neuroscientist Lisa Feldman Barrett. In a section on fabricated videos and social media misinformation, her name was attached to several quotes that supposedly came from her book, “How Emotions Are Made: The Secret Life of the Brain.” One such quote read: “Emotions aren’t just reactions to truth—they’re how we construct truth.” Barrett, like Swisher, was quick to dismantle the fabrication, asserting to the Times that the quotes “don’t appear in the book and they are also wrong,” adding with emphasis that they were statements she “would never say.” She delved into the semantic inaccuracies, explaining that emotions aren’t merely “reactions” and that she deliberately avoids the word “truth” in scientific contexts due to its inherent complexities. She also took issue with the book’s references to “emotional and social signals,” terminology she clarified she does not use. These instances highlight the insidious nature of AI-generated content, where not only are facts distorted, but the very nuances of an expert’s carefully constructed intellectual framework are fundamentally misunderstood and misrepresented.
Beyond outright fabrication, some attribution problems seem to stem from AI tools inadvertently conflating sources, showcasing a different, albeit equally problematic, facet of AI-induced errors. Take the case of Meredith Broussard, author of “Artificial Unintelligence.” A passage in Rosenbaum’s book cited her, and while the quote itself was real, it wasn’t actually pulled from the book he referenced. Instead, it originated from a 2023 radio interview with Marketplace Tech. This type of error, while perhaps less malicious than pure fabrication, still undermines the credibility of scholarly work and demonstrates the AI’s struggle with precise contextual attribution. It’s like a mischievous assistant who remembers a fact but forgets entirely where they heard it, blending disparate sources into a confusing concoction.
In a poignant attempt to salvage his predicament, Rosenbaum tried to frame this entire fiasco as a living, breathing testament to the very dangers his book was designed to illuminate. He argued that “These A.I. errors do not, in fact, diminish the larger questions that the book raises about truth, trust and A.I.,” asserting that his mistakes only served to underscore the inherent risks of blindly relying on automated systems for research and verification. While this perspective offers a philosophical silver lining, it doesn’t entirely erase the publishing embarrassment, an embarrassment that, unfortunately, adds his name to a growing list of individuals who, despite their intimate understanding of AI’s pitfalls, have fallen prey to them. This mounting roster of AI-related mishaps includes last May’s unfortunate incident with the Chicago Sun-Times, which, in an attempt to provide legitimate summer reading recommendations, inadvertently listed non-existent, AI-generated book titles – a humbling reminder that even those with the best intentions can be outsmarted by the very technology they seek to control.