AI and Journalism: are newsrooms catching the next tech wave?
First you went online. Then you got into social media. You are now ‘digital first’. So, is now the time to start thinking about Artificial Intelligence?
I have spent the last few months gathering the thoughts of more than 70 digital leaders in news organisations around the world on how they think Artificial Intelligence is going to change their newsrooms and the journalism industry as a whole.
You can now read the full report, but it is clear that this is a major technological development. Not as dramatic as the Great Leap Online but similar to the rush to get on social media. At a time when news organisations are already busy making radical changes to their business models the danger is that once again, journalism will be left chasing after the change instead of meeting the challenge.
[This report is a version of an article that first appeared in Inpublishing Magazine]
First, the ‘good’ news for those worried about a robot takeover. There is no such thing as the kind of genuinely intelligent artificial intelligence displayed by the sinister computer Hal in 2001 Space Odyssey. Yes, there are programmes that can beat grand masters at board games and Chinese AI-simulated TV newsreaders. But these, like most ‘AI’ currently used in journalism, are essentially programmes that can be trained to ‘learn’ from data sets to make predictions, classify information and organise text or images. It is better described as machine learning and its uses include relatively simple tools such as automation and bots.
The working definition for our research was disarmingly broad and simple-sounding:
“A collection of ideas, technologies, and techniques that relate to a computer system’s capacity to perform tasks normally requiring human intelligence”
A lot of that has been with us for some time. You are using ‘AI’ when you search on Google or scroll through your Twitter of Facebook feed. The advertising you see programmed to chase you around the web is driven by algorithms that seek to target content to suit you.
Our research gathered an impressive list of current uses already deployed by newsrooms from America to Africa. They covered every aspect of the publishing flow: newsgathering, content creation, distribution, personalisation and marketing or monetisation. This ubiquity is key. You might be aware of particular examples of AI used in news such as RADAR, an automated programme created by URBS Media with PA that churned out 50,000 local stories in its first three months of operation. But the reality is that AI can be used across just about every aspect of news production.
Newsrooms told us about AI-driven practices that made their work more efficient and effective such as Trint that provides automated transcriptions of audio interviews. It is the kind of technology that frees up time for the journalist to concentrate on creating added value. Another AI widget monitored the images on a newspaper’s website and helped the picture editor ensure a better gender balance of photos. The reader wouldn’t notice but the journalism had been improved.
But perhaps the biggest area of development was around getting the right content to the audience and so increasing engagement. This could involve creating completely new automated products such as Quartz Bot Studio’s intelligent chatbot, an AI-based tool for serving journalism in an instant messaging, conversational context. Or it could involve finessing the user/content interface.
The Times’s JAMES (“Journey Automated Messaging for Higher Engagement through Self-Learning”) uses data to get to know the habits, interests, and preferences of readers, acting as a ‘digital butler’ serving them the articles that they really want to read. For a news organisation like The Times whose business model depends on its subscription paywall system this is vital to optimise reader’s experience. That reduces churn and maximises attention.
The New York Times took this further with its ‘Project Feels’ that used sentiment analysis of content to understand and predict the emotional impact of its articles on readers. It used this to serve appropriate advertising alongside them in a personalised way.
The list of use cases was wide-ranging: automated moderation of comments; tracing trending topics; fact-checking; neural networks for photo-tagging; video editing; packaging for voice; recommended reading; Freedom of Information requests; and, of course, lots of data journalism from automated sports or financial results to deep mining of complex data sets.
What difference does it make?
But what difference did it make to those newsrooms that have started to put these ideas into practice? The answer is both ‘a lot’ and ‘less than you’d think’. Generally, most newsrooms are taking a step-by-step approach and so a lot of energy and effort is being spent on both preparation and fitting these new tools or systems into the existing production structures. Smaller newsrooms struggle to find the AI-skilled staff who know enough about editorial to take single-uses to the next level.
The vast majority of newsrooms were enthusiastic. AI saved them time, improved their journalism and enhanced the relationship with the customer. But scaling that up and spreading it across the whole spectrum of production needs a strategic confidence that isn’t always there yet. It is having an impact now but the significant effects are still 3–5 years away.
It is still relatively early days for this technology. Our report discovered a lot of enthusiasm for AI in the future. But there was a big divide between those news organisations that can do this in a ‘holistic’ way across the organisation and those who are working on specific tools or projects. The well-funded Wall Street Journal, for example, has now put AI at the heart of newsroom re-organisation with an expanded R&D team, to build a ‘data science and audience engagement powerhouse’. Another newspaper described how AI-related activities went right across its operations:
“Recommendation of related articles. Robot journalism. Personalization of the newsfeed. Lookalike audiences based on user data to increase CPMs. Predictive analytics to optimize news curation. Speech-to-text services to increase our editors productivity. Churn prediction and prediction of the propensity to subscribe. Tagging / entity recognition. Spell check”
Yet it might be a single-function that is truly transformative. It seems we might be just a year or two away from tools that can produce convincing, accurate and natural live text and speech translation. That could be a major breakthrough for smaller specialist or national newsrooms currently limited by the barriers of language. It would also give a major boost to reporting on international issues at a time when the need to know about what’s happening in the world.
Publishing as an industry should pay attention to what other sectors are doing with AI. Banking, retail, finance, law, health, security and, of course, social media have all been empowered to greater efficiencies and effectiveness. The big organisations have the muscle to make the most of this at scale, but it has also helped fuel an explosion of smaller start-ups filling niches or disrupting legacy markets. We found evidence of this in the news industry, as nimble newcomers fill gaps. Some established brands are investing through incubators or funding of innovative third-party companies. But overall media is a relatively small sector in the world of AI and needs to work harder to keep up with the pace.
Obstacles to progress
There are some real obstacles to progress. Newsroom culture is still fearful of the ‘robots taking our jobs’ or removing the creative or human element from journalism. There is a huge knowledge gap throughout the news teams and management. The technology appears complex and journalism-friendly products or systems are often still in development stage. And resources. While the health, security, and retail industries have billions to throw at research and development and experimentation, cash-strapped media usually needs support.
That is leading to some interesting efforts at collaboration — with Universities for example — and the growth of intermediary companies who can sell off-the-shelf AI products across the sector. And, of course, the tech giants who have by far the most AI expertise and research resources are keen to make sure that they are central to this new market and the data it creates. As publishers know from past efforts, cuddling up to the digital behemoths is not always a pain-free experience.
I was impressed by the commitment and ingenuity of these AI-interested newsrooms. They can see that this technology can have huge structural benefits. They are also conscious of the hazards of algorithmic bias and getting over-excited about ‘bright shiny’ tech hype. They nearly all stressed that AI will augment, not replace the journalist. It has to fit with the business logic and editorial values of the brand. AI is in many ways an invisible technology. If it is going to improve the quality and sustainability of journalism we are all going to have to get clued-up and plugged-in and soon. Transparency will be key. It is in the interest of the journalist and the news consumer that we open up the black box of AI and work out together with the technologists how it might shape the future of publishing.
This article was written by Charlie Beckett, who is a professor in the Media and Communications Department at the London School of Economics and Political Science (LSE). He is also the director of Polis, the LSE’s international journalism think-tank. He is leading the Journalism AI project that reports this November. The project is funded by the Google News Initiative