The other day I asked ChatGPT about a political moment I’ll never forget: “What was the 47 percent story about?”
asserted
story → ask → What
Its answer:
The “47 percent story” refers to a controversial comment made by Mitt Romney, who was the Republican nominee for President of the United States in the 2012 election.
asserted
who → refer → election
In a private fundraising event that was secretly recorded and later leaked to the media, Romney stated that 47 percent of Americans are dependent on government support, believe they are victims entitled to benefits, and consequently, would vote for then-President Barack Obama.
asserted
they → record → Obama
That’s accurate as far as it goes.
asserted
it → ’ → ?
But there’s one glaring omission: any reference to Mother Jones, which uncovered the story that reshaped that election cycle.
asserted
that → ’ → cycle
The summary contained no links or citations to our reporting.
asserted
summary → contain → reporting
You’d have thought it all sprang fully formed from the AI’s, uh, head.
asserted
it → think → head
I asked some follow-up questions and ChatGPT spit out lots more details about the story, including lengthy quotes from the video.
asserted
ChatGPT → ask → video
The whole thing took maybe 20 seconds.
asserted
thing → take → seconds
By contrast, it took David Corn, our Washington bureau chief, months to get the scoop.
asserted
it → take → scoop
Digging into the candidate’s background in the hedge-fund world, David tracked speeches in which Romney had bragged about investments that exported American jobs overseas.
asserted
that → dig → jobs
Eventually, that led him to the secret video, at which point David and a team in our newsroom spent weeks verifying the content, reporting out additional context, and working out ways to present it to the public while protecting his sources.
asserted
David → lead → sources
OpenAI, of course, didn’t do any reporting to create its summary.
asserted
OpenAI → do → summary
They simply hoovered up the text of the story to help train ChatGPT to talk like a human.
asserted
They → hoover → human
We know this because in its early versions, OpenAI listed the sources of its data (something it has since clammed up about), and in those datasets, we found tens of thousands of our stories.
asserted
we → know → stories
OpenAI never asked for permission to use our work, nor did they offer to pay a licensing fee.
asserted
they → ask → fee
Not then, when they were building their business, and not now when they are getting ready for an initial public offering that may end up around $1 trillion.
uncertain
that → build → trillion
Nor, so far as we know, did OpenAI ask for permission or a license from the countless other publishers, writers, creators, or internet posters (including, perhaps, you) whose intellectual property they ingested.
asserted
they → know → property
It’s not that one of the most valuable companies in the world can’t afford to pay.
asserted
one → ’ → world
They have money for offices, electricity, free lunches, stratospheric pay packages, data centers, and many other things.
asserted
They → have → offices
OpenAI claims they can use the content for free, under a legal concept called fair use, because they are transforming it into something new, like a musician who samples a track to create a fresh composition.
uncertain
who → claim → composition
But unlike an artist honoring a predecessor, OpenAI “transformed” the information in such a way that no one would ever know where it came from, by stripping copyright and author information as they processed it.
asserted
they → honor → it
That’s like stealing your car and then grinding off the VIN number (only with millions or billions of cars).
asserted
That → ’ → cars
Maybe OpenAI thought that publishers, especially nonprofit ones like us, would not be able to put up a fight.
asserted
publishers → think → fight
Our budget, after all, is something like one-half of one percent of Sam Altman’s net worth, and their legal team—let’s just say you can’t toss a golf ball in San Francisco or Washington without hitting one of them.
asserted
you → let → them
And OpenAI has Donald Trump on its side too: This week, the administration went to the trouble of filing a brief in the lawsuit, an unusual step for the government to take at such an early stage in the process.
asserted
government → have → process
The brief supports OpenAI, saying it’s a vital national interest for AI companies not to have to pay for content (with a dig about how that would just be a subsidy for “old mainstream media companies” anyway).
asserted
that → support → companies
Given Trump’s history with paying people, that tracks.
asserted
that → give → people
Just as not all communities are rolling over for data centers (despite Trump ordering them to “let data reign”), some publishers are fighting back, and we are one of them.
asserted
we → roll → them
For the last two years we’ve been involved in a lawsuit asking OpenAI to respect our copyright, and September 4 marks a crucial deadline, when both sides are asking the court to decide the case ahead of going to trial.
asserted
sides → involve → trial
So it seems a good time to unpack what’s going on, because this is about a lot more than journalism.
asserted
this → seem → journalism
As a society, we’re starting to grapple with the fact that tech moguls are building things they don’t understand and can’t control.
asserted
they → start → things
Much as when humans first split the atom, we are dealing with technology that can be both immensely useful and extremely dangerous, and right now it doesn’t seem as if our leaders, from corporate CEOs to politicians, are willing to grapple with those risks.
asserted
leaders → split → risks
As Matteo Wong and Charlie Warzel write in The Atlantic, “it can feel like no individual person, organization, or government is really steering the technology’s development, not because machines have displaced humans but because so many humans have already chosen to step aside.”
asserted
humans → write → humans
One reason it’s so frustrating to watch this happen is that while the tech is new, the playbook is incredibly familiar.
asserted
playbook → ’ → ?
From Uber and Airbnb steamrolling communities when they rolled out their products, to Flock putting all of us under surveillance, we’ve seen this movie before: Tech company does what it wants, takes what it needs, and asks permission later.
asserted
it → steamroll → permission
Back when Facebook and Google were trying to pull in users by offering news, they’d occasionally hand out checks to publishers—paying them to pivot to video, appear in a promoted “news tab,” or be part of a fact-checking program.
asserted
they → try → program
Inevitably, they would hand-pick big, commercial publishers to partner with, further centralizing the information ecosystem.
asserted
they → pick → ecosystem
(In the end, of course, the joke was on the publishers when the tech giants pulled back on news and threw open the doors to disinformation.
asserted
giants → pull → disinformation
And now the AI companies are playing that same game.
asserted
companies → play → game
…and 19 more, not listed.
