AI revolutionizes journalism

Robots have infiltrated the news and are automating the journalistic process, but it’s not quite time to start planning a human revolution just yet.

In a time of shrinking editorial staffs and an increasing amount of things to cover, newsrooms have turned to artificial intelligence and robo-journalists to shoulder some of the burden.

Read two financial articles based on the same earnings report published on the same day as an example - one was written using artificial intelligence, one by a human journalist.

"The online retailer posted revenue of $37.96 billion in the period, topping Street forecasts. Ten analysts surveyed by Zacks expected $37.21 billion," one reads.

"Revenues of $38 billion. Analysts were expecting $37.2 billion. Amazon posted sales of $30.4 billion in the second quarter last year," says the other.

For formulaic and data-driven earnings stories, it’s not easy to tell which was generated by an algorithm. The reader sees the same information, but the time it saved that reporter is critical in today’s constant news cycle. (For the record, the first excerpt was written by AI.)

The Associated Press partnered with startup Automated Insights and began rolling out earnings report articles in 2014, and the move to automation is shaking up how journalists do their jobs.

"The key point is AI should not be used to displace people from their jobs - that's not what's happening," says Francesco Marconi, who co-leads global automation and artificial intelligence strategy at the AP. "It actually helps free up resources and people’s time to focus on other things."

As reporters focus less on writing basic earnings stories, they can spend more time looking at the bigger picture, says Ted Merz, global head of news product at Bloomberg. While the computer handles the initial news - getting basic information including revenue or share price out - the reporter can already be talking to sources, researching the context, or comparing earnings reports from competing companies for a larger piece.

Automated stories aren’t limited to financials either. The AP also uses Automated Insights’ program Wordsmith for other formulaic news such as sports. The Los Angeles Times built QuakeBot to automatically generate news stories about earthquakes, which pulls information from the U.S. Geological Survey seconds after the earthquake occurs.

The bots, however, still need help from a human. A journalist or programmer needs to write a specific template story for the bot to populate. Marconi compared the story template to the fill-in-the-blank game Mad Libs.

An example earnings template might read: [Company name] posted [revenue] in [quarter], a [percent increase/decrease] compared to [last quarter/last year].

The AI reads a press release for the matching data and writes the story. This is all done with natural language processing, in which a computer is trained to read and comprehend human writing and produce its own writing understandable to humans, having been trained on thousands of writing samples.

Finally, the story is read by an editor to check for mistakes and published - corporate earnings reports don’t see human eyes until they are ready to go live.

However, as Sylvie Barak, media consultant and senior director at FTI Consulting and former tech journalist, points out: "The press release became irrelevant a long time ago and AI makes it even more irrelevant. AI is not going to be bullshitted. It doesn't care how well you present your facts - it only cares about the facts."

As algorithms become more common, journalists need a basic understanding of how they work - to write a template for the bot, a reporter needs to know what it can and cannot do.

Justin Hendrix, executive director of NYC Media Lab, a group that partners New York universities, tech companies, and media organizations to drive innovation in media, says the economics of the media industry will change as technology is embedded in newsrooms.

"Journalists’ job functions will change," Hendrix explains. "People will spend their time training computers to do things that formerly only humans could, like sifting through documents, looking at datasets, or reviewing footage."

AI isn’t only taking news writing off a reporter’s plate. Media organizations are also developing tools to help journalists keep up with the constant flow of news online and on social media.

"Reporters and PR professionals are similar in the sense they're looking for information and need to process it in some way," Merz says. "We're just at the beginning of information overload. People think they’re processing lots of information now, but it will be nothing compared to the years to come."

Bloomberg has been using machine learning-enabled feeds to monitor breaking news for about four years, Merz says. On Twitter alone, he explains, 500,000 tweets out of the 500 million sent each day are financially relevant to its reporters, so Bloomberg needed a way to sift through them and find the important ones.

Merz speculates that one day press releases will undergo a similar AI monitoring process, being read first by a computer to determine whether it’s newsworthy.

"Imagine the stream of press releases coming in every day," he says. "If we build a tool that allows our newsroom to get a better indication or a score saying this press release is more likely to be significant, they can focus on that because otherwise everything is treated the same. These tools allow reporters and editors to prioritize time and focus on bigger stories."

Hendrix says NYC Media Lab works with tech startups that have developed algorithms to edit video footage into a news clip, tag and caption photos, or determine the mood of a person on video.

"The awesome thing about AI is it doesn't just help with the shrinking newsroom," says Barak. "You free up your top journalists to do actual journalism while AI does the grunt work. AI is going to free up so much value and actually protect journalists’ jobs."

And PR people must double down on the relationship-building part of their job and "make themselves much better at giving context and extra information on top of bare bones data," notes Barak.

All of this work sounds impressive, but experts stress it is only the beginning. As AI continues to advance, especially in the realms of language understanding and generation, its uses in the newsroom will expand.

"Tech adoption follows the same patterns whether you’re talking about social media, AI, or VR," Marconi says. "First, people are hesitant or have concerns, then they experiment and finding best practices, and eventually they deploy tools at scale. Most news organizations have a social media strategy, they will start having an AI strategy. In 10 years it will just be called strategy."

AI IN THE NEWS