How to make your newsroom AI-ready
First steps in creating an AI roadmap for news organizations
Artificial Intelligence is transforming present-day journalism and is set to become the future backbone of innovative newsrooms. But there is a risk of an innovation gap, as not every news organization will have access to these new automation technologies. For this reason, in my explorations as a John. S. Knight (JSK) Journalism Fellow at Stanford, I’ve been seeking ideas on how to create an AI roadmap for newsrooms.
A Medium post I published in December (“10 things about AI every newsroom should know,”) drew a lot of interest from journalists and technologists. The question many of them asked, no matter their country or news organization, is “Where do we begin?” Artificial Intelligence can seem a bit scary or daunting to many journalists; there’s a real need for more guidance to introduce AI in a newsroom setting.
So I reached out to people at a few key news organizations that have already integrated AI into their news operations and are looking to do more. I interviewed innovation leaders at The Associated Press, The Washington Post and the Finnish public broadcaster Yle. The following recommendations could be a start in helping news organizations take the first steps toward using AI in their newsrooms.
1. Pick a small project to introduce AI
How will you power your AI? This a crucial question when you want to introduce AI for the first time in your newsroom. The availability of clean and consistent data will decide what kind of pilot project you can start. Sports and business news have been the most frequent first use cases for newsrooms, followed by election results.
The Associated Press and The Washington Post were among the first news organizations to automate stories about sports and business. “There are two important criteria for why we choose to automate stories,” explains Lisa Gibbs, director of news partnerships at The Associated Press. “We have to be able to demonstrate that this automated content will reach new audiences or enter new markets, and that we save a lot of time by pursuing automation.”
Emilio Garcia-Ruiz is The Washington Post managing editor who oversees the newsroom’s development, implementation and execution of digital strategy. “Our people were very encouraging and impressed when we introduced Heliograf, our in-house program to create online stories on high school sports. This was content that we weren’t creating yet,” he says. “Keeping a journalist in the loop is essential. Because it is one thing for Heliograf to write a cohesive sentence using a template, but the final result should also be publishable.”
Last year, the Finnish public broadcaster Yle introduced their robo-journalist Voitto, the Finnish word for victory. It uses machine learning to write articles about the results of municipal elections and ice hockey games. But Voitto is not just for the journalists at Yle. Integrated into the Yle NewsWatch app, Voitto is also a personal smart news assistant for public users, recommending interesting articles on their mobile devices.
“We gave a lot of thought on how to present Voitto, how it looks and feels,” says Jarno Koponen, the head of AI at Yle. “It was our choice to represent the robo-journalist as a character, not like the Terminator or an anonymous machine. We developed Voitto in close cooperation with our journalists, to see how it could help them. Now they are very happy: Voitto is augmenting their work and taking over the more mechanical tasks, while the journalists can focus more on the human work: analysis and storytelling.”
2. Work with an in-house AI team or with start-ups
For the development of new AI applications, news organizations have to choose between in-house expertise and outsourcing the work, perhaps to a start-up partner or university. Two factors are decisive: time and money. Outsourcing is usually faster, while hiring data scientists and machine learning experts can cost a lot. In the long term, building in-house expertise will pay off and can give your news organization a more competitive advantage.
The Finnish broadcaster Yle decided to develop its AI-applications in-house. “It should be an effort of the whole company,” said Koponen. “We now have a multidisciplinary team of about 30 to 40 people in our news lab, depending on the project. A combination of people with a strong journalism background, analytical people, experts in machine learning and in user experience.”
The Washington Post, owned by Amazon founder Jeff Bezos, chooses to develop everything in-house and invested heavily in a team of almost 250 tech people. According to Garcia-Ruiz, the data scientists and developers provide a better understanding of their readers. “They help us with what type of content we should bring and what types of stories we should place where. The big advantage of having this in-house team is that you can work together with journalists who can provide the correct journalistic context.”
But not every newsroom has the resources to hire expensive data scientists. “We’ve found it more efficient to work with a start-up, instead of building applications in-house,” says the AP’s Gibbs. “This is much faster, and in our business we have to act fast. The start-ups have the expertise that we don’t have in the newsroom. At AP we’ve developed a playbook to identify startups or university departments that have specific expertise. We cut deals with them because we can provide them with important training data from our archives. We really started to re-appreciate our archives, thanks to AI. Even for smaller newsrooms, these archives are a valuable asset in this new era.”
3. Train your journalists in AI
AI will transform several jobs in the newsroom. Giving journalists the knowledge to adapt to new roles in the AI era is essential. News organizations can decide to hire consultants and AI-experts to educate employees. But thanks to rise of MOOCS (massive open online courses) and webinars from such organizations as the Knight Center, Coursera and YouTube, it has become more cost effective to train employees in new skills such as elements of AI.
“At Yle we are trying to create practices in different departments of the newsroom,” says Koponen. “As this thinking becomes more commonplace, so our training will evolve. This is an ongoing process. It’s important to have good examples and we always demonstrate our products to the journalists. We try to be present at the daily newsroom briefings to see where we can help the journalists understand what AI can do to detect and cover stories.”
Gibbs believes that the training should start at universities — and there’s an important step to take first. “Journalism professors ask me if they should teach their students about AI. But I believe that first and foremost they have to teach them about data. Understanding data and how you can leverage it is key. In our newsroom we now have an automation editor in-house who is able to code the templates. As the technology will become easier and cheaper to use, I think we’ll see plenty of new jobs.”
4. Build an AI strategy
After successful execution of the initial AI projects and some basic training, journalists will have a better understanding of how AI can augment their work. Newsroom leaders should be able to identify the areas where AI can create added value and draw the outlines of a coherent AI strategy.
At Yle, they believe a good strategy is a matter of survival. “This strategy should always start from the user perspective,” said Koponen. We focus on 3 levels: augment the user citizen, augment the user journalist and augment the user newsroom. We now build tools for each of these groups. We always ask ourselves first what kind of user experience we want to create. After this we check what kind of technology we need to apply to accomplish this. It is not tech first, but user first.”
Gibbs agrees that having an AI strategy is critical. “As a news organization, we must leverage AI and other automation technologies to do our job better and more efficiently,” she said. “This is inherent of the business we’re in, because we’re short of time and resources.” As this world gets more and more sophisticated, Gibbs believes journalists shouldn’t only be prepared to understand the technology, but they also need to be prepared to fight the battle against bad actors who could use AI for ill. “Therefore, it absolutely makes sense to have at least one designated AI-expert in your newsroom, who keeps tabs on the use cases in our business.”
According to Garcia-Ruiz, there is a recent shift in general journalism strategy to focus less on quantity, but more on quality. “If the emerging business model is subscription based, you have to ask yourself what is the best AI journalistic use that could lead to more subscriptions? In local news, you have to find out what the local instances are that can add value to your publication. Earnings companies from local companies, for example, or local sports and local court records. Focus on areas that have robust datasets. But don’t overuse it. Find things that work and double down on that success.”
Compared to a large publisher like The Washington Post, smaller publishers have many other hard challenges to face right now before they will get to AI, Garcia-Ruiz acknowledges. “They just don’t have the resources, and they certainly don’t have access to developers. They will have no other choice than to partner up with the Google News Initiative, for example, to build certain tools.”
5. Communicate about the use of AI
Last but not least, as AI will affect all key stakeholders of a news organization, it is important to implement a strategic communication plan. This should not only help reduce reluctance to adopt AI, both internally and externally; it should also address concerns that jobs will be automated and replaced by AI.
Gibbs believes good communication is the key to success with AI. “Always think of how you are going to onboard the whole newsroom,” she said. “At the AP we have very open communication about our goals. We involve reporters and editors in the team. It is essential that they are being integrated in the process.” She said she understands that the introduction of any new tool means a change of workflow for journalists. “Some people are more flexible than others to adopt new tools. Therefore, you must provide multiple opportunities for feedback. We use Slack, for example, for the testing of our latest news detection tool.”
“In the end, our job is all about storytelling,” concludes Koponen. “You must try and create a direct connection with your customer. Without this, you’re in danger to become obsolete as a media company. The great thing is that we’ll see a whole new form of storytelling emerging, thanks to these new technologies.
A roadmap can help news organizations to implement artificial intelligence. Is your newsroom already experimenting with AI and do you want to share best practices? Please reach out to me at tom.vandeweghe@stanford.edu .
