MAGAZINE PUBLISHES SERIOUS ERRORS IN FIRST AI-GENERATED HEALTH ARTICLE

Home / Artificial Intelligence / MAGAZINE PUBLISHES SERIOUS ERRORS IN FIRST AI-GENERATED HEALTH ARTICLE

MAGAZINE PUBLISHES SERIOUS ERRORS IN FIRST AI-GENERATED HEALTH ARTICLE

From Neoscope, 02-09-23

When the publisher of Sports Illustrated and Men’s Journal announced last week that its magazines would start to publish AI-generated articles, its CEO assured readers that the practice wouldn’t result in a decline in quality.

“It’s not about ‘crank out AI content and do as much as you can,’” Arena Group CEO Ross Levinsohn told the Wall Street Journal. “Google will penalize you for that and more isn’t better; better is better.”

The stakes were high, and not just because the small handful of publications that have already tried to replace human writers with AI systems, like CNET, have caught fierce backlash over errors and plagiarism found in the AI’s work. In a key way, Arena Group’s foray was even more perilous: it was using AI to provide readers with advice about the particularly consequential topics of health and medicine.

Take the very first article the bot published in Men’s Journal, which carries the title “What All Men Should Know About Low Testosterone,” and the human-sounding byline “Men’s Fitness Editors.” The story issued a cornucopia of medical claims, nutrition and lifestyle advice, and even suggested a specific medical treatment in the form of testosterone replacement therapy, all aimed at readers looking for guidance on a serious health issue.

Like most AI-generated content, the article was written with the confident authority of an actual expert. It sported academic-looking citations, and a disclosure at the top lent extra credibility by assuring readers that it had been “reviewed and fact-checked by our editorial team.” 

But on closer inspection, the whole thing fell apart. Bradley Anawalt, the chief of medicine at the University of Washington Medical Center who has held leadership positions at the Endocrine Society, reviewed the article and told Futurism that it contained persistent factual mistakes and mischaracterizations of medical science that provide readers with a profoundly warped understanding of health issues.

“This article has many inaccuracies and falsehoods,” he said. “It lacks many of the nuances that are crucial to understand normal male health.”

Anawalt pointed to 18 specific errors he identified in the article. Some were flagrantly wrong about basic medical topics, like equating low blood testosterone with hypogonadism, a more expansive medical term. Others claimed sweeping links between diet, testosterone levels, and psychological symptoms that Anawalt says just aren’t supported by data.

“There is just enough proximity to the scientific evidence and literature to have the ring of truth,” Anawalt said, “but there are many false and misleading notes.”

All told, scrutiny by a human expert makes Arena’s foray into AI-generated content look less like the empowering new set of tools that Arena’s CEO Levinsohn described in the Wall Street Journal — and more like a cynical cash grab aimed at churning out dirt-cheap content designed to capture readers on Google, even if it means providing them with abjectly false medical information.

It’s worth noting that Men’s Journal has already been through the wringer. In early 2020, it was reported that the magazine’s owner at the time had laid off its entire editorial staff. Arena Group only bought it a few months ago, in December 2022, and its staff page currently lists only five employees, two of whom are part-time. 

In other words, AI seems to give media executives a new instrument to hasten the already-ravaged journalism industry’s relentless race to the bottom, as operators like Arena use it to speedily pump substandard content into the husks of once-loved publications — quality be damned.

When Futurism reached out to Arena and several representatives of Men’s Journal about the issues, a cascade of changes started to appear in the AI-generated testosterone article, correcting various errors pointed out by Anawalt until the piece was almost unrecognizable. 

A new message also appeared at the bottom, referring to several of Anawalt’s critiques: “The original version of this story described testosterone replacement therapy as using ‘synthetic hormones’ and stated poor nutrition as one of the most common causes of low T, which are inaccurate.”

But numerous additional changes were made in the article that weren’t mentioned in that note, such as removing a claim about the health risks of milk that Anawalt had flagged. And many more of his criticisms went unaddressed.

After the testosterone article had been thoroughly rewritten, an Arena spokesperson sent a brief statement.

“Along with many media companies, we are exploring various AI software that can be additive to our editorial teams’ workflow,” it read. “An initial pilot collating previously published archival content demonstrated the potential of AI coupled with editors and writers. These early experiments are a work in progress. Based on these learnings and ongoing monitoring, we will continue to refine our use of these tools as part of our workflow, which has been and will always be anchored in editorial oversight.”

Asked whether Arena thought it was responsible to publish health advice that needs to be almost completely rewritten, whether it was confident in the accuracy of the rest of the bot’s articles, and whether it had plans to keep publishing the AI content, the spokesperson sent a terse and difficult-to-interpret reply.

“Confident, yes, but no immediate plans,” they wrote, clarifying that the company was “confident in the articles” and “continuing to monitor how we enhance our publishers reaching consumers in new ways.”

Providing flawed health advice produced by AI could run the risk of invoking serious consequences from Google, the de facto intermediary between publishers like Arena and potential readers. Just this week, Google issued new guidelines for publishers about using AI to generate content, saying it would tolerate the practice but that there would be extra emphasis on the reliability of search results involving “health, civic, or financial information.”

The bungling deployment of AI tech at Men’s Journal mirrors what we’ve seen elsewhere as publishers have struggled to incorporate AI in a way that provides value to readers.

Many were alarmed to discover in January, for instance, that the prominent tech news site CNET had been quietly publishing dozens of AI-generated articles about personal finance. After near-universal condemnation, further reporting by Futurism found that the AI-generated articles were loaded with basic errors and substantially plagiarized.

After our reporting, CNET — along with sister sites Bankrate and CreditCards.com, also owned by publisher Red Ventures — paused the publication of the AI content for a review. CNET eventually admitted that more than half the articles the bot had published contained either factual mistakes, plagiarism, or both.

In spite of it all, CNET pledged to forge ahead with the AI-generated articles — as soon as the negative press had subsided, that is. 

Like Men’s Journal, CNET had fallen on hard times before turning to AI content. It was once a titan of tech journalism, but staff there told The Verge recently that since its acquisition by Red Ventures in 2020, its editorial independence has been undermined by pressure to treat advertisers favorably.

All told, the shoddiness of CNET’s AI efforts seems to have presaged those at Men’s Journal. Instead of learning from CNET’s mistakes and carefully probing the work generated by the AI for any signs of weakness, Men’s Journal — or Arena Group puppeteering its plundered carapace, as the case may be — seems to have fallen into the same trap: seeing that AI systems like ChatGPT can easily churn out confident-sounding content, and mistaking that conviction for accuracy.

Are the debacles at CNET and Men’s Journal signs of things to come? Maybe, but they don’t have to be. 

It’s true that if the public — not to mention regulators and power players like Google — allow jaded corporations to steamroll the internet with content spammed out by AI they didn’t bother to understand, it does risk drowning out good journalism and advice with irresponsible, mass-produced misinformation.

But even if those publishers have no shame, an outcry against the practice of polluting the net with dumpster-loads of AI-written garbage could have tangible effects. It could dissuade more brands from jumping on the bandwagon in the future, for one. It’s also possible that advertisers will balk at funding worthless AI-generated content. Or, perhaps most meaningfully, it could push Google to consider real consequences for publications that use AI to publish garbage at scale.

Leave a Reply

Your email address will not be published.