Politico’s Rushed Adoption Of Half-Cooked ‘AI’ Continues To Go Terribly

(Mis)Uses of Technology

from the I'm-sorry-I-can't-do-that,-Dave dept

We’ve noted repeatedly how early attempts to integrate “AI” into journalism have proven to be a comical mess, resulting in no shortage of shoddy product, dangerous falsehoods, and plagiarism. It’s thanks in large part to the incompetent executives at many large media companies, who see AI primarily as a way to cut corners, assault unionized labor, and automate lazy and mindless ad engagement clickbait.

The folks rushing to implement half-cooked AI at places like Red Ventures (CNET) or G/O Media aren’t competent managers to begin with. Now they’re integrating “AI” with zero interest in whether it actually works or if it undermines product quality. They’re also often doing it without telling staffers what’s happening, revealing a widespread disdain for their own employees.

This sort of behavior has been particularly problematic at the DC gossip rag Politico, where ownership continues to implement “AI” systems that don’t really work all that well, constantly introduce new errors human editors have to correct, and engage in “behavior” that violates editorial standards. All without the input and knowledge of actual journalists or editors.

This rundown on the Politico AI mess by Brian Merchant is worth a read. He documents how in one instance, Politico Editorial staff were told just an hour before a new AI product was introduced, given zero chance to ask questions about how it would work, why it was there, or why it was being launched. The tech then immediately proceeded to make a bunch of embarrassing, rookie mistakes:

“The AI promptly generated a post that misspelled Kamala Harris’s mother’s name. The entry was taken down without comment or correction from an editor, in apparent violation of Politico’s editorial standards. Weeks later, Politico’s management deployed the AI tool again, this time in an even higher-profile setting: The vice presidential debate between JD Vance and Tim Walz. The feature again trampled editorial guidelines, this time transcribing verbatim Vance’s comments about “illegal immigrants”—a term that Politico writers are not allowed to use, and editors are not supposed to publish.”

Politico management introduced another AI “report builder” for premium Politico PRO subscribers. It’s supposed to offer a breakdown of existing Politico reporter analysis of complicated topics. Apparently the “AI” sucks at doing that, as well:

“It’s wholly behind the paywall, but when we have asked it things, it’s giving us back some pretty glaring errors,” [Politico journalist Ariel] Wittenberg says. “I asked it about ‘The Impact of President Biden’s Oil Policies,’ and it wrote me a whole page-and-a half thing, and every single policy it mentioned was a policy of Trump’s. And it cited real stories at the very bottom, from our members, the implication being that if someone is reading this, and it’s erroneous, not only does our AI not know the difference between Biden’s policies and Trump’s, but maybe the authors of the cited articles didn’t, either.”

Great stuff! This reflects the same experiences with other major media outlets that expect LLMs to genuinely understand their own outputs. You’ll recall that Apple had to pull its AI news synopses system because the AI routinely introduced glaring mistakes; VC “AI” marketing has many execs believing we’re just a few steps from full sentience, when these systems are still struggling with the very basics.

We’ve seen the same nonsense over at Microsoft, which was so keen to leverage automation to lower labor costs and glom onto ad engagement that they rushed to implement AI across the entirety of their MSN website, never really showing much concern for the fact the automation routinely produced false garbage. Google’s search automation efforts have been just as sloppy and reckless.

Again, the “automation” in this instance is also a direct reflection of the quality of Politico’s ownership, who likes Trump and actively embraces lazy “both sides” journalism that buries the truth in pursuit of fake ideological objectivity. It’s not any sort of coincidence that owners who don’t care about journalism, transparency, labor, or factual reality introduce broken tools that harm all four.

In this case, the introduction of the new automation wasn’t just rushed and lazy, it was in direct violation of the editorial union’s contract struck just last year. So union employees have since been battling with Politico via arbitration:

A lot of affluent media owners have also just completely drunk the marketing Kool-Aid on modern “AI” as just a few skips away from full sentience; they genuinely believe modern LLMs are more capable than they actually are. And they’re soo keen on using this emerging technology to cut corners, “save money” and undermine and replace pesky unionized labor, they’re blind to the fact it often doesn’t work.

The CEO of Politico Owner Axel Springer, Mathias Döpfner, recently introduced a company wide mandate that every single employee in the organization has to not only use AI, but consistently file reports justifying why they don’t. This sort of stuff goes way past useful technological adoption and teeters into delusional religion, and the technology becomes a window into very ordinary human failures.

Filed Under: ai, automation, errors, journalism, labor, llms, mathias dopfner, media, politico, unions
Companies: axel springer, politico