Robots can save local journalism, but will they make it more biased?

Local journalism has taken a hit in recent years. In the UK, at least 198 local papers closed between 2005 and 2018, a decline that prime minister Theresa May warned is a threat to democracy.

One solution for stopping the decline of local journalism is to use robot reporters.

The Press Association (PA) -- the British news agency founded in 1868 -- has recently started to automate their news writing. Newspapers are requesting more and more coverage from their areas these days, says PA's editor Pete Clifton. “What Radar allows us to do, with a small number of additional journalists, is more local stories.”

PA's Radar project aims to create up to 30,000 stories a month by algorithm. A partnership of PA and the data start-up Urbs Media, the project is funded via a €700,000 grant from Google's Digital News Initiative. Radar's robot journalist uses publicly available data sets to produce stories on a wide array of topics, from birth rates and child obesity rates to crime figures. Its first stories have made it into print in papers such as the Norwich Evening News and the Cambridge Independent.

The rise of the robots

PA isn't the first organisation to try robots. Major news agencies in Europe have adopted automation. Even before PA announced their project, research I conducted for the Reuters Institute for the Study of Journalism in Oxford found that at least seven news agencies in European countries were using robot news writing. My 2017 report on automation in news agencies concluded that agencies do not use robots to replace journalists, but to expand the coverage of stories they didn't cover before.

Local news is particularly daunting for automated journalism. In many places, local data is available but not reported on. In several cases mentioned in my report, automation provides a boost to local news. Two examples come from Nordic countries. Norwegian news agency NTB has created a robot journalist that writes local football stories for as many as 20,000 league games played in Norway each year. The Swedish news agency TT launched a widget that compares figures for school performance and real estate prices with the national average and writes a short text summary for readers.

In the case of PA, the Radar project covers statistics on a local level that before were only reported on a national basis.

Making local news scale

Automation allows reporting to be massively scaled. It also raises new questions about doing local journalism. In the world we once knew, each local paper covered stories in their own way. Now everyone gets news from the same template, with the same set of angles. What does this mean for public discourse?

To make better sense of PA's news writing algorithms, I tried to approach automated stories like any grouser complaining to his local paper -- I read them looking for flaws. This is what I found:

A robot story in the West Midlands paper Express and Star was headlined ‘Majority of new mothers in Wolverhampton are unmarried’. The story opens by saying that 56.5% of children are born to parents neither in a marriage or civil partnership, but later in the story it notes that 77% of children are born into households with two parents.

The robot's decision to make the story about half of the children born 'out of wedlock' (as the 1950s phrase goes) gives it a socially conservative spin. It is also potentially misleading, as almost 4 in 5 children can count on the care of at least two people in the household.

An automated piece in the Romford Recorder notes that only 16% of local children from disadvantaged backgrounds make it to university. While the story mentions national statistics, it does little to tie in other local data, such as cuts to the council budget. This is surprising, as the Recorder had reported a planned budget cut by a staggering 98% earlier, which will affect local services and likely worsen the life of disadvantaged children. While PA's template allows for broader national context, the robot clearly knows little of local circumstances.

The Royal United Hospital in Bath apologised after a robot story noted that doctors cancelled four urgent and potentially life saving operations in October 2017, a finding drawn from statistics of the UK's National Health Service. This shows that robots not only can break a story, but force local institutions to react and thus create political realities with their finding.

This leads me to a couple of conclusions. One, robot reporting will never be entirely neutral and its templates will reflect the political bias of its creators. That is, because in picking one angle over the other when looking at data, the robot can't help it but choose a side to a story. Hence, the story about childbirth will somehow not only report data, but also presents a social viewpoint on it.

In addition, data is neither neutral or context-free. As the Romford case shows, statistics in and of themselves can obscure the larger picture when they fail to mention other data relevant to a story.