Can AI do journalism?

The robot reporter debate ... On-again, off-again ... And cheaters never prosper.

The past few decades have felt like a neverending bloodbath in the journalism industry.

News outlets big and small have been gutted by lost revenue. Every few months, an outlet announces another round of newsroom layoffs. Trust in news organizations is down, and the people in charge don’t seem to know how to fix it all.

Now, there’s a shiny new tool that promises to “do more with less” and maybe bring the industry back to its former glory.

If you’ve been reading our new weekly, the A.I. Agenda, you probably know where this is headed. AI has a lot of potential to make newsrooms faster, stronger and smarter. AI can research and crunch numbers faster than humans and ease the mind-numbing labor of transcribing public meetings.

There’s a real risk that AI-generated news will become a cheap, ubiquitous replacement for real journalism — but can journalists’ core responsibility to ask tough questions, pick up on nuance and push back against power be outsourced to a machine?

You may notice that our art intern, ChatGPT, is finally learning how to spell. Six months ago, it would have spelled newsletter four different and incorrect ways.

Since generative AI came on the scene several years ago, the news industry has seen its share of scandals:

• Last week, the LA Times launched its AI “Insights” app that summarizes news stories. Within days, it caused quite a controversy when it summarized an article about the Ku Klux Klan with very fuzzy language that sounded a lot like AI was sympathizing with the hate group.

• Sports Illustrated was busted two years ago for running stories and photos that were generated by AI. Even the authors of the stories were made up by AI and used AI-generated profile photos. The publication denied that any stories were written by AI, although they removed the fake authors from their website.

• A Wyoming reporter was caught last August using AI to make up quotes from the state governor and a local prosecutor.

• In December, Apple launched an AI feature that helped summarize news articles for users, which had to be recalled because it started manufacturing stories like claiming that Luigi Mangione had shot himself.

That’s a lot to take in. So today we’re exploring AI from the lens of a newsroom — including our own.

But first, let’s start by looking at all the places AI is currently used in journalism. AI is already a tireless research assistant, crunching massive datasets to uncover trends, drafting initial versions of stories (because even journalists love a good rough draft), whipping up visuals and videos and translating content.

Image source

In fact, a Thomson Reuters Foundation survey revealed that over 81% of 200 journalists across 70 countries are already using AI tools somewhere in their editorial pipeline.

We’ve been thinking a lot about AI lately, in part because we use a lot of AI features within Skywolf, our legislation-tracking service. An AI model summarizes all the bills introduced at the Arizona Legislature, and a veteran human bill summarizer reviews and improves that summary before it’s published. Then she trains the model so it can do a better job next time.

So when we show you a Skywolf bill summary, that’s already based on an AI-generated, human-reviewed snippet of work.

But where are the redlines? Where’s the balance between utilizing AI for better, faster, more accurate journalism, and just getting lazy and handing our responsibilities over to the robots?

The journalism industry hasn’t quite answered those questions yet.

Big News Ethics

While major publications have taken steps to address the ethics of AI, many of their policies are surprisingly vague and often feel like placeholders rather than concrete guidelines.

And many of those policies were written years ago, when AI tools like ChatGPT were first coming onto the scene.

The Washington Post, for example, touts transparency about its AI use, but stops short of outlining exactly how or when disclosure occurs. It’s AI policy says:

• We are transparent about how and when we use AI

• We allow AI to suggest related content, to translate between languages, or to sift through pages of documents or thousands of images

• We will not employ AI to generate images, video or visual works that purport to represent reality without disclosing its use

• We believe in the value of our intellectual property and will protect the integrity of our work wherever it is used

The Associated Press treads cautiously, acknowledging AI's potential, yet hedging with terms like “experiment with caution“ and “benefit the mission.” The AP explicitly discourages using tools like ChatGPT for publishing, but there's no clarity on oversight or enforcement.

• AP staff may experiment with ChatGPT with caution, they do not use it to create publishable content

• Any output from a generative AI tool should be treated as unvetted source material

• AP does not allow the use of generative AI to add or subtract any elements from photos, or videos

• AP urges staff not to put any sensitive materials into AI tools

The New York Times and others proudly emphasize human expertise, essentially declaring "machines can't replace judgment," which sounds comforting but sidesteps the real issues of when and how to use AI.

• Generative AI can assist our journalists in uncovering the truth

• The expertise and judgment of our journalists are competitive advantages that machines simply can’t match

• The first principles of journalism should apply just as forcefully when machines are involved

Tech publication Wired offers one of the more detailed policies, which explicitly prohibits AI-generated — and AI-edited — content.

• No using AI generated photos to support stock photos

• No publishing stories with text generated by AI

• No publishing text edited by AI

• AI can be used to suggest headlines or social media posts

• AI can be used to research topics

• Artists may use AI to spark inspiration

This week, the Agenda set out to draft our own AI ethics policy. But we quickly found that, like many of the big national outlets, we’re still not sure where we land.

So we’re trying a different approach…

To read about how the Agendaverse is approaching AI, and watch us argue about it in real-time, click over to the A.I. Agenda.

Today’s edition of the A.I Agenda explores how the Agenda will use — and not use — AI tools as we continue to innovate local, independent journalism.

And while you’re there, don’t forget to fill out the A.I. Agenda survey to tell us how you think journalists should use AI — and how you’re using AI in your life.

Those who foray into the glorious world of municipal budgeting may encounter three types of budgets for any given fiscal year.

The adopted budget is very common and can be found in the wild throughout the year, and usually remains on a municipal web site for at least several years.

The amended budget is somewhat more rare, as it usually signals some kind of change to the city/town/county/state budget after it was adopted.

A grant for a specific department that comes in the middle of the budget year, for example, would turn an adopted budget into an amended budget.

Both examples are great sources of information, providing a broad overview of each department and where the revenue comes from and how the money is spent.¹

Our favorite budget, by far, is the audited statement. Usually performed by an outside, third-party auditing firm, these budgets show how much was actually spent during the fiscal year (and whether there were any reporting irregularities) - but are usually available three to six months after the fiscal year ends.

It’s a Catch 22. Budgets for the City of Tucson are being built right now for the FY25/26 fiscal year, with those documents being adopted by the city no later than this June.

These budgets will rely on forecasts and supplemental requests made by each department, not actual spending. We don’t get a clear picture on how much they will have spent until the audited statements are available this fall.

We like to use audited statements to see how they compare to the adopted budgets. It’s a quick and dirty way to see who didn’t spend their entire budget (delayed programs/capital projects, and vacancies are often good culprits) and those who overspent on their budgets.

As we move forward with our budget bytes series, we will be using all of these documents to help explain the city’s budget, and give you a better understanding of how your tax dollars are spent.

It’s not too late to vote on Proposition 414, but please don’t put your ballot in the mail.

You can drop off your ballot at the following locations:

• 310 N Commerce Park Loop

• 7200 E Tanque Verde Road

• 2160 N. 6th Avenue

• 8155 E Poinciana Drive

• 101 W Irvington Rd

• 900 S. Randolph Way

• 800 E. 12th Street

Yesterday, a lot of Agenda subscribers voted in our poll about Prop 414, so let’s keep it going for another day. After everything that’s been said and reported about Prop 414, what did you end up deciding?

Signatures in: It looks like Miranda Schubert will be back on Tucson ballots this summer. Schubert announced she had submitted 1,151 petition signatures to the Tucson City Clerk's office on Monday night. The sheer number of signatures - the maximum allowed by the city - all but guarantees her place on the Democratic primary ballot. A successful petition challenge would require convincing a judge that more than half her signatures were invalid. Four other candidates - three Democrats and one Republican - have filed statements of interest for the Ward 6 seat. They have until April 7 to turn in their signatures.

Tariff whiplash: Produce distributors in Southern Arizona and the growers they work with in Mexico are trying to make sense of the on-again, off-again tariffs from the Trump administration, the Arizona Daily Star’s Emily Bregel reports. Cross-border commerce is a bipartisan issue in Southern Arizona and both Democratic Gov. Katie Hobbs and Republican U.S. Rep. Juan Ciscomani said they would bring the produce industry’s concerns to federal officials.

“It’s absolutely awful, anxiety-ridden. You can’t future plan. You can’t even next-week plan,” Nogales seed distributor Sabrina Hallman said.

Taking a hard look: Arizona Luminaria and Arizona Public Media are offering a free screening of the mini-documentary they put together on deaths in the Pima County jail. The screening will be March 18 at the University of Arizona Modern Languages Building, Room 350 at 5:30 p.m. There also will be a panel discussion after the screening.

Spending choices: Tucson Sentinel columnist Blake Morlock had some choice words for state officials who insist on “stat shaming” schools, and the conservative think tanks that criticize low-income schools for how they spend their money. It’s relevant at the moment as the state audits schools to determine how much money is spent in classrooms. Meanwhile, a nationwide study of students showed low-income and minority students in the Tucson Unified School District are having a hard time closing the achievement gap, the Star’s Jessica Votipka reports.

We mentioned a Wyoming reporter who got caught using AI to generate fake quotes from officials. But we didn’t tell you about the incident that finally ended his scheme.

The reporter ran this quote about the comedian Larry the Cable Guy serving as grand marshal for a local parade.

“The 2024 Cody Stampede Parade promises to be an unforgettable celebration of American independence, led by one of comedy’s most beloved figures. This structure ensures that the most critical information is presented first, making it easier for readers to grasp the main points quickly.”

The second part of that quote is just a description of what we call the “inverted pyramid,” a simple, effective way to write news stories. It has nothing to do with the story, but the AI spat it out and the reporter didn’t catch it.

1. We could make the case for even more transparency here - we’d love a really clear explanation on how much each department spends on travel in every city/town/county we cover.