Generative AI brings wrongness at scale

“For all its promise, generative AI can get more wrong, faster — and with greater apparent certitude and less transparency — than any innovation in recent memory.”

The giant robot in the corner would like a word, or 500. It’s been standing there quietly for a while, helping out with mundane tasks — sorting and filing, looking up info and the like — generally staying out of the way unless called on. You might not have even noticed it. But you will.

Earlier this year, the text-to-image generation tools Stable Diffusion, Midjourney, and DALL-E 2 made impressive public debuts. You’ve no doubt seen the results in your social media feeds. And just this month, OpenAI released ChatGPT, a chatbot based on its unsettlingly smart GPT-3 large language model — quickly driving home both the promise and peril of generative AI.

Made with DALL-E 2: “Illustration of a robot sitting awkwardly at a conference table in a newsroom alongside several editors, who look concerned.” (A human illustrator would no doubt have done this better.)

It’s hard to overstate the disruptive potential of the machine-creative revolution we are witnessing — though some are clearly trying: “The death of artistry.” “The end of high-school English” (and the college essay, too, evidently). We now have non-developers producing functional code and a children’s book written and illustrated entirely by machine.

Of course, AI tools have been all over our industry for a while now: We’ve used them for transcription, translation, grammar checking, content classification, named entity extraction, image recognition and auto-cropping, content personalization and revenue optimization — among other specific purposes.

But emerging use cases made possible by generative tools — including text and image creation and text summarization — will broaden the scope of AI’s impact on our work.

I don’t imagine we’ll see GPT-3-produced copy in the pages of The New York Times in 2023, but it’s likely we’ll turn to machines for some previously unthinkable creative tasks. As we do, we will hopefully reflect on the risks.

Even the best generative AI tools are only as good as their training, and they are trained with data from today’s messy, inequitable, factually challenged world, so bias and inaccuracy are inevitable. Because their models are black boxes, it is impossible to know how much bad information finds its way into any of them.

But consider this: More than 80% of the weighted total of training data for GPT-3 comes from pages on the open web — including, for example, crawls of outbound links from Reddit posts — where problematic content abounds.

Add to that the tools’ disconcerting habit of obscuring sources and presenting wildly incorrect information with the same cheery confidence it applies to accurate answers, and you have high potential for misinformation (to say nothing of the dangers of deliberate misuse).

Will these tools get better? Undoubtedly. We may be in something of an uncanny valley stage, and who knows how long that will last?

Right now, though, my take is this: For all its promise, generative AI can get more wrong, faster — and with greater apparent certitude and less transparency — than any innovation in recent memory. It will be tempting to deploy these tools liberally, and we know that some black-hat SEOs will be unable to pass up the opportunity to publish thousands of seemingly high-quality articles with zero human oversight. But the possible uses will always be more numerous than the advisable ones.

(And what will happen to model training, fact-checking, and general user experience when more and more of the information on the open web is produced by AIs? Will the web become one big AI echo chamber? OpenAI is already trying to build watermarking into GPT-3 to facilitate the detection of AI-generated text, but some experts believe this is a losing battle.)

Applying these powerful tools surgically to narrowly-defined use cases, while keeping humans in the loop and providing needed sourcing transparency (and credit!), will enable us to wield them for good.

But if we fail to build the necessary checks on AI’s creations, then the likelihood of students passing off robot-written text as their own will be the least of our worries.

Eric Ulken is a product director at Gannett.

The giant robot in the corner would like a word, or 500. It’s been standing there quietly for a while, helping out with mundane tasks — sorting and filing, looking up info and the like — generally staying out of the way unless called on. You might not have even noticed it. But you will.

Earlier this year, the text-to-image generation tools Stable Diffusion, Midjourney, and DALL-E 2 made impressive public debuts. You’ve no doubt seen the results in your social media feeds. And just this month, OpenAI released ChatGPT, a chatbot based on its unsettlingly smart GPT-3 large language model — quickly driving home both the promise and peril of generative AI.

Made with DALL-E 2: “Illustration of a robot sitting awkwardly at a conference table in a newsroom alongside several editors, who look concerned.” (A human illustrator would no doubt have done this better.)

It’s hard to overstate the disruptive potential of the machine-creative revolution we are witnessing — though some are clearly trying: “The death of artistry.” “The end of high-school English” (and the college essay, too, evidently). We now have non-developers producing functional code and a children’s book written and illustrated entirely by machine.

Of course, AI tools have been all over our industry for a while now: We’ve used them for transcription, translation, grammar checking, content classification, named entity extraction, image recognition and auto-cropping, content personalization and revenue optimization — among other specific purposes.

But emerging use cases made possible by generative tools — including text and image creation and text summarization — will broaden the scope of AI’s impact on our work.

I don’t imagine we’ll see GPT-3-produced copy in the pages of The New York Times in 2023, but it’s likely we’ll turn to machines for some previously unthinkable creative tasks. As we do, we will hopefully reflect on the risks.

Even the best generative AI tools are only as good as their training, and they are trained with data from today’s messy, inequitable, factually challenged world, so bias and inaccuracy are inevitable. Because their models are black boxes, it is impossible to know how much bad information finds its way into any of them.

But consider this: More than 80% of the weighted total of training data for GPT-3 comes from pages on the open web — including, for example, crawls of outbound links from Reddit posts — where problematic content abounds.

Add to that the tools’ disconcerting habit of obscuring sources and presenting wildly incorrect information with the same cheery confidence it applies to accurate answers, and you have high potential for misinformation (to say nothing of the dangers of deliberate misuse).

Will these tools get better? Undoubtedly. We may be in something of an uncanny valley stage, and who knows how long that will last?

Right now, though, my take is this: For all its promise, generative AI can get more wrong, faster — and with greater apparent certitude and less transparency — than any innovation in recent memory. It will be tempting to deploy these tools liberally, and we know that some black-hat SEOs will be unable to pass up the opportunity to publish thousands of seemingly high-quality articles with zero human oversight. But the possible uses will always be more numerous than the advisable ones.

(And what will happen to model training, fact-checking, and general user experience when more and more of the information on the open web is produced by AIs? Will the web become one big AI echo chamber? OpenAI is already trying to build watermarking into GPT-3 to facilitate the detection of AI-generated text, but some experts believe this is a losing battle.)

Applying these powerful tools surgically to narrowly-defined use cases, while keeping humans in the loop and providing needed sourcing transparency (and credit!), will enable us to wield them for good.

But if we fail to build the necessary checks on AI’s creations, then the likelihood of students passing off robot-written text as their own will be the least of our worries.

Eric Ulken is a product director at Gannett.

Joshua P. Darr   Local to live, wire to wither

Burt Herman   The year AI truly arrives — and with it the reckoning

Moreno Cruz Osório   Brazilian journalism turns wounds into action

Jessica Maddox   Journalists keep getting manipulated by internet culture

Bill Grueskin   Local news will come to rely on AI

Cassandra Etienne   Local news fellowships will help fight newsroom inequities

Janelle Salanga   Journalists work from a place of harm reduction

Pia Frey   Publishers start polling their users at scale

Nicholas Diakopoulos   Journalists productively harness generative AI tools

Leezel Tanglao   Community partnerships drive better reporting

Anika Anand   Independent news businesses lead the way on healthy work cultures

Sue Robinson   Engagement journalism will have to confront a tougher reality

Matt Rasnic   More newsroom workers turn to organized labor

Sam Guzik   AI will start fact-checking. We may not like the results.

Peter Sterne   AI enters the newsroom

Laura E. Davis   The year we embrace the robots — and ourselves

Kaitlyn Wells   We’ll prioritize media literacy for children

Sue Schardt   Toward a new poetics of journalism

Amethyst J. Davis   The slight of the great contraction

Delano Massey   The industry shakes its imposter syndrome

John Davidow   A year of intergenerational learning

Jacob L. Nelson   Despite it all, people will still want to be journalists

Barbara Raab   More journalism funders will take more risks

Rodney Gibbs   Recalibrating how we work apart

Jennifer Brandel   AI couldn’t care less. Journalists will care more. 

Eric Nuzum   A focus on people instead of power

Ariel Zirulnick   Journalism doubles down on user needs

Sarabeth Berman   Nonprofit local news shows that it can scale

Molly de Aguiar and Mandy Van Deven   Narrative change trend brings new money to journalism

Megan Lucero and Shirish Kulkarni   The future of journalism is not you

Joni Deutsch   Podcast collaboration — not competition — breeds excellence

Gordon Crovitz   The year advertisers stop funding misinformation

Andrew Losowsky   Journalism realizes the replacement for Twitter is not a new Twitter

Laxmi Parthasarathy   Unlocking the silent demand for international journalism

S. Mitra Kalita   “Everything sucks. Good luck to you.”

Priyanjana Bengani   Partisan local news networks will collaborate

Ståle Grut   Your newsroom experiences a Midjourney-gate, too

Elizabeth Bramson-Boudreau   More of the same

Mariana Moura Santos   A woman who speaks is a woman who changes the world

Rachel Glickhouse   Humanizing newsrooms will be a badge of honor

Surya Mattu   Data journalists learn from photojournalists

A.J. Bauer   Covering the right wrong

Mario García   More newsrooms go mobile-first

Kavya Sukumar   Belling the cat: The rise of independent fact-checking at scale

Jody Brannon   We’ll embrace policy remedies

Jaden Amos   TikTok personality journalists continue to rise

Jessica Clark   Open discourse retrenches

Francesco Zaffarano   There is no end of “social media”

Mael Vallejo   More threats to press freedom across the Americas

Josh Schwartz   The AI spammers are coming

Alexandra Svokos   Working harder to reach audiences where they are

Taylor Lorenz   The “creator economy” will be astroturfed

Eric Thurm   Journalists think of themselves as workers

Daniel Trielli   Trust in news will continue to fall. Just look at Brazil.

Anthony Nadler   Confronting media gerrymandering

Martina Efeyini   Talk to Gen Z. They’re the experts of Gen Z.

Raney Aronson-Rath   Journalists will band together to fight intimidation

Juleyka Lantigua   Newsrooms recognize women of color as the canaries in the coal mine

Paul Cheung   More news organizations will realize they are in the business of impact, not eyeballs

Julia Beizer   News fatigue shows us a clear path forward

J. Siguru Wahutu   American journalism reckons with its colonialist tendencies

Susan Chira   Equipping local journalism

Alexandra Borchardt   The year of the climate journalism strategy

Cindy Royal   Yes, journalists should learn to code, but…

Jonas Kaiser   Rejecting the “free speech” frame

Dannagal G. Young   Stop rewarding elite performances of identity threat

Gina Chua   The traditional story structure gets deconstructed

Eric Ulken   Generative AI brings wrongness at scale

Brian Moritz   Rebuilding the news bundle

Christoph Mergerson   The rot at the core of the news business

Gabe Schneider   Well-funded journalism leaders stop making disparate pay

Julia Angwin   Democracies will get serious about saving journalism

Khushbu Shah   Global reporting will suffer

Ryan Kellett   Airline-like loyalty programs try to tie down news readers

Mar Cabra   The inevitable mental health revolution

Ryan Gantz   “I’m sorry, but I’m a large language model”

Kirstin McCudden   We’ll codify protection of journalism and newsgathering

Sue Cross   Thinking and acting collectively to save the news

Michael W. Wagner   The backlash against pro-democracy reporting is coming

Jennifer Choi and Jonathan Jackson   Funders finally bet on next-generation news entrepreneurs

Alex Sujong Laughlin   Credit where it’s due

Ayala Panievsky   It’s time for PR for journalism

Zizi Papacharissi   Platforms are over

Stefanie Murray   The year U.S. media stops screwing around and becomes pro-democracy

Kerri Hoffman   Podcasting goes local

Upasna Gautam   Technology that performs at the speed of news

Masuma Ahuja   Journalism starts working for and with its communities

Larry Ryckman   We’ll work together with our competitors

Victor Pickard   The year journalism and capitalism finally divorce

AX Mina   Journalism in a time of permacrisis

Anna Nirmala   News organizations get new structures

Esther Kezia Thorpe   Subscription pressures force product innovation

Nicholas Jackson   There will be launches — and we’ll keep doing the work

Jim VandeHei   There is no “peak newsletter”

Nikki Usher   This is the year of the RSS reader. (Really!)

Christina Shih   Shared values move from nice-to-haves to essentials

Tre'vell Anderson   Continued culpability in anti-trans campaigns

Hillary Frey   Death to the labor-intensive memo for prospective hires

Jesse Holcomb   Buffeted, whipped, bullied, pulled

Sarah Stonbely   Growth in public funding for news and information at the state and local levels

Mauricio Cabrera   It’s no longer about audiences, it’s about communities

Amy Schmitz Weiss   Journalism education faces a crossroads

Jenna Weiss-Berman   The economic downturn benefits the podcasting industry. (No, really!)

Simon Galperin   Philanthropy stops investing in corporate media

Emily Nonko   Incarcerated reporters get more bylines

Andrew Donohue   We’ll find out whether journalism can, indeed, save democracy

Sam Gregory   Synthetic media forces us to understand how media gets made

Richard Tofel   The press might get better at vetting presidential candidates

Eric Holthaus   As social media fragments, marginalized voices gain more power

Elite Truong   In platform collapse, an opportunity for community

Joe Amditis   AI throws a lifeline to local publishers

Basile Simon   Towards supporting criminal accountability

Joanne McNeil   Facebook and the media kiss and make up

David Skok   Renewed interest in human-powered reporting

Alex Perry   New paths to transparency without Twitter

Brian Stelter   Finding new ways to reach news avoiders

Karina Montoya   More reporters on the antitrust beat

Cari Nazeer and Emily Goligoski   News organizations step up their support for caregivers

Jakob Moll   Journalism startups will think beyond English

Sarah Marshall   A web channel strategy won’t be enough

Dana Lacey   Tech will screw publishers over

Shanté Cosme   The answer to “quiet quitting” is radical empathy

Kathy Lu   We need emotionally agile newsroom leaders

Parker Molloy   We’ll reach new heights of moral panic

Walter Frick   Journalists wake up to the power of prediction markets

Ben Werdmuller   The internet is up for grabs again

David Cohn   AI made this prediction

Janet Haven   ChatGPT and the future of trust 

Danielle K. Brown and Kathleen Searles   DEI efforts must consider mental health and online abuse

Peter Bale   Rising costs force more digital innovation

Sumi Aggarwal   Smart newsrooms will prioritize board development

Michael Schudson   Journalism gets more and more difficult

Errin Haines   Journalists on the campaign trail mend trust with the public

Kaitlin C. Miller   Harassment in journalism won’t get better, but we’ll talk about it more openly

Tamar Charney   Flux is the new stability

Nicholas Thompson   The year AI actually changes the media business

Felicitas Carrique and Becca Aaronson   News product goes from trend to standard

Al Lucca   Digital news design gets interesting again

Sarah Alvarez   Dream bigger or lose out

Don Day   The news about the news is bad. I’m optimistic.

Emma Carew Grovum   The year to resist forgetting about diversity

Ryan Nave   Citizen journalism, but make it equitable

Jarrad Henderson   Video editing will help people understand the media they consume

Mary Walter-Brown and Tristan Loper   Mission-driven metrics become our North Star

Lisa Heyamoto   The independent news industry gets a roadmap to sustainability

Anita Varma   Journalism prioritizes the basic need for survival

Dominic-Madori Davis   Everyone finally realizes the need for diverse voices in tech reporting

Tim Carmody   Newsletter writers need a new ethics

Johannes Klingebiel   The innovation team, R.I.P.

Valérie Bélair-Gagnon   Well-being will become a core tenet of journalism

Snigdha Sur   Newsrooms get nimble in a recession

Jim Friedlich   Local journalism steps up to the challenge of civic coverage

Cory Bergman   The AI content flood

Doris Truong   Workers demand to be paid what the job is worth

Alan Henry   A reckoning with why trust in news is so low

Bill Adair   The year of the fact-check (no, really!)

Wilson Liévano   Diaspora journalism takes the next step