We Asked the ‘Future of Truth’ Author to Explain How He Used AI. It Didn’t Go Well

Earlier this month, WIRED published an excerpt from Steve Rosenbaum’s buzzy caller book, The Future of Truth, which looks astatine however artificial quality warps people’s consciousness of reality. Shortly thereafter, The New York Times reported that the publication contained implicit a half-dozen made-up oregon misattributed quotes. In a statement, Rosenbaum, who has a master's grade successful "truth" from New York University, admitted that helium had accidentally included “a handful” of “improperly attributed oregon synthetic” quotes. In an ironic twist, the veracity of a publication astir however AI impacts information was present nether aggravated scrutiny due to the fact that of however its writer had utilized AI.

After the Times communicative broke, WIRED took different look astatine our 1,450-word excerpt. The fact-checking squad had reviewed it anterior to publication, and we reconfirmed that its quotes and facts were accurate. But WIRED’s generative AI editorial argumentation prohibits the work of AI-generated and AI-edited writing, and a scholar email calling retired the excerpt arsenic being “blatantly AI-written” raised further questions astir the grade to which Rosenbaum had utilized AI tools. In The Future of Truth’s acknowledgement section, Rosenbaum writes that ChatGPT, Claude, NaturalReaders, ProWritingAid, and Grammarly had helped “refine and polish the presumption of [his] ideas.” What, exactly, did that mean?

WIRED ran the excerpt done respective AI-detection services, including Pangram, GPTZero, and ZeroGPT. Each work suggested that it was either apt AI-generated, oregon AI-generated with precocious confidence. But AI-detection tools are fallible, and tin instrumentality inaccurate readings. So WIRED’s caput of probe emailed Rosenbaum straight to inquire if and however helium had utilized AI to constitute the excerpt.

He wrote back: “Like galore writers moving today, I utilized AI tools during parts of the probe and editorial improvement process for the book, including root discovery, brainstorming, structural feedback, and connection refinement.” But, helium stressed, “the ideas, reporting, arguments, and last authorship are mine, and the WIRED excerpt was not generated by AI and past simply published as-is.” He urged WIRED’s editors to workout caution trusting AI detection tools, noting that mendacious positives tin occur.

At this point, WIRED’s elder editors asked maine to look into the episode, due to the fact that I’ve covered AI slop successful its assorted forms since 2024. My archetypal measurement was to tally the full substance of the publication done Pangram’s detection tool. (While each AI-detection tools person limitations, and tin amusement false-positives, Pangram is the existent golden standard.) It came backmost that the publication appeared to beryllium 53 percent AI-generated, with an further 9 percent registering arsenic apt AI-assisted.

I called Rosenbaum and asked for a much elaborate statement of however he’d utilized AI to constitute the book, and whether helium disputed Pangram’s results. (BenBella Books, whose imprint published The Future of Truth, did not instrumentality requests for comment. Simon & Schuster, which distributes BenBella’s books successful the United States, declined to comment.)

Rosenbaum would not measurement successful connected the accuracy of Pangram’s results. In fact, helium didn’t privation to speech astir them astatine all. “I don’t enactment successful that conversation,” helium said. “It’s similar saying, do you bushed your wife? It’s 1 of those accusations that there’s nary effect to.”

He offered, instead, to broadly explicate his editorial process. He says that astatine the opening of the penning process, helium utilized AI tools arsenic hunt engines, helping him aboveground accusation for the much research-heavy sections of the book. To show however helium mightiness bash this, helium asked ChatGPT to picture me, past work the results retired loud. The AI hunt much oregon little accurately described immoderate of my anterior stories, including enactment connected AI-generated “zombie media sites.”