Why are news organisations interested in Artificial Intelligence? How is AI being used? What are the challenges ahead as news companies implement, drive, and structure for this technology?
Tomchak’s presentation covered four key areas that represent the emerging opportunities in AI for news publishers:
- Assisted newsgathering.
- More efficient production.
- Better distribution.
- Commercial opportunities.
Why news organisations are interested in AI
Tomchak started the Webinar by sharing that 72% of news publishers are currently experimenting with Artificial Intelligence, according to the RISJ Digital Leaders Survey. The key areas of interest in the technology centre around:
- Delivering more relevant and engaging content.
- Reducing cost through intelligent automation.
- Producing new revenue streams.
But it’s clear these aren’t practical solutions yet, Tomchak said. “We aren’t sure where it will go yet. Machine learning isn’t nearly good enough yet to not have humans involved,or to scale.”
Assisted newsgathering and identifying news stories
This technology has been around for at least a decade, Tomchak said, and is based on using data to find stories. A number of organisations are already using AI to help with their data. There are actually a number of smaller companies that have been developed to use AI to do things like trawling for big stories and using automatic language translation to speed up the process.
They also use data to identify clusters that show patterns to help journalists and editors understand where the stories might be. “Dataminr does the final piece quite well,” he said, “turning this into useful information for journalists.”
When it comes to verification, it’s not just about spotting fake news. “It means: Is the source of content credible? Is the date and time of the video correct? Is the location correct? AP, in particular, has been at the forefront of this in my opinion — in developing robot information-gathering technology.”
Media companies also need to make sure that once they’ve gotten the information gleaned from AI and machine learning, they can actually make it useful as it comes into the building, Tomchak added: “You have to make sure the information, once it’s brought in, remains useful and easy to search so that it continues to serve you.”
AI has been used in a number of areas for production already. Tomchak shared several examples:
- Subbing of stories: checking for grammar and spelling, as well as accessibility, tone, and popularity.
- Intelligent automation: ALTO is a voiceover tool for repurposing video content into multiple languages using text-to-speech voice synthesis.
- On-screen robo journalists: China has created AI newsreaders, and there are anime reporters in Japan. These are presented as “journalists with no ego, who work 24 hours a day and never ask for a pay raise.”
- ReporterMate from The Guardian: Creates stories that are simple and digestible, almost listicle-type articles that are pulled from public information. Importantly, the stories are disclosed as being automated.
In some of these cases there are authenticity issues, Tomchak said. “In other words, they’re a little gimmicky.”
There is also editing that needs to happen, though that should happen across all areas of the newsroom. “It frees journalists up to do more actual journalism.”
One of the challenges is the possible dangers of robojournalism, as was demonstrated in a fake Mark Zuckerberg video.
“We need to make sure that robojournalism and AI combined are used in the right way, which is not to spread misinformation,” Tomchak said.
Distribution has all kinds of opportunities when it comes to AI, Tomchak shared with INMA members. Personalisation is a big part of this.
The areas that people are particularly interested in centre around filters, making sure that they pull topics that the reader has a history of reading, popularity, etc. From the editorial side, factors to be taken into account include editorial value, date that the story was published, and editorial tags.
“Don’t just take what [the readers have] seen in the past, but use patterns to offer things your reader might be interested in,” he said. “There’s a danger with personalisation that newsrooms try to do things all the time and lose sight of their audiences.”
He shared how Schibsted handles personalisation with an easy tool, moving popular stories to the best positions: “This is an interesting example of how easy it can be.”
This leads to the question: To what extent do we create filter bubbles by our own hand? “What kind of bias is built into the decision making process around personalisation?”
The more algorithms are used in everyday work, the more teams must build diversity into the system, he said. “If you don’t build a team of people who are diverse, you will get filter bubbles that are biased.”
Tomchak admitted he doesn’t have all the answers to how media companies can monetise and capitalise on AI yet, though there are several areas of promise:
- Audience growth: doing more to grow audiences for advertisers and sponsors.
- Syndication: the Internet of things and sharing revenue with platforms.
- User acquisition and data revenue: premium content and paywalls.
- Affiliates: product sales.
- Further commercial diversification: events, technology, creative solutions, and other AI products.
One specific example Tomchak shared was the case study of how Lexus built an AI to write an ad. “It’s all about intuition and emotion, along with the technology,” Tomchak said. “I found this fascinating the first time I saw it. Thinking about the audience, thinking about the emotion of the story, and using it in a way — in theory at least — for a machine to write something that will tug at my heart strings and maybe sell me something.”
When it comes to the actual ad that resulted, however — called Driven by Intuition — Tomchak says he was left flat. “But the idea is a good one, the concept is a good one. This sense of allowing machines to tell the stories that are coming through media is very interesting.”
It is vital that media organisations work as a group to make sure they combat things like fake news as the industry moves forward with AI, Tomchak said. He identified the main challenges presented with AI and news:
- Clarifying how AI can help journalism — how it can help tell better and more relevant stories.
- Building up skills and systems to understand how to use it well.
- Being transparent with consumers about the role AI technology has played in the journalism.
- Explaining the benefits to staff in a way that doesn’t pit human against machine.
“I think there are lots of opportunities to use this in ways that will help journalism. But if we don’t bring people with us in the newsroom, they will see it as a threat rather than an opportunity.”
Ultimately, AI is about innovation and thinking about that as we move forward is important, Tomchak concluded.
INMA: What type of time frame do you think it will be for AI to become the norm?
Tomchak: I think from a technology perspective, it’s something that could happen very quickly, 12 to 24 months. But the technology is not really the issue; it’s more about the journalism. And that should set the pace. We shouldn’t throw away the quality of journalism in a rush to technology. We need to figure out how to use it properly — and that will come from human beings, not from the technology itself.
INMA: For which company size would you say AI starts to scale or payback?
Tomchak: We are very small. We’re not tiny, but not to the scale of some of our big competitors. What I’m trying to say is that any size is fine to start doing this stuff. But you have to do it well. At the moment, there are so few people who are talking to platforms about this that you can be ahead of the curve. If I was starting a company from scratch right now, I would start with AI from the beginning.
INMA: What’s the role for journalists when it comes to setting the tone in AI content?
Tomchak: That’s what journalists do, I would suggest. What the AI does is learn from the journalists, so they set the tone. Editors set the tone. People know what we stand for, a reflection of the city. Ultimately in the news gathering we do, people will see what’s important to us and what’s not. All of that tone comes from the leadership and the journalists on the ground.
That is clearly a complicated mix, and if I was to ask people on the floor what was our tone, we would get different answers. But AI can go in and look back at everything we’ve written and see the choices we have made (text, video, audio, images) to then compile something that can be sent to an editor — done in a way that’s in our tone — and then the editor can sub it.
INMA: How easy is it to automate personalisation?
Tomchak: Shibsted’s tool is actually very easy, as we saw from their example. It’s drag and drop. Our CMS is probably like most, it’s fairly open and easy to plug in APIs for different things. I don’t think these things are particularly hard; it’s about prioritisation. Do you want your home page editor to have something else to do? Our time is limited, our resources are limited. I suggest you look at your audience and your core values and then look at some of the tools that are out there and try them.