The Rise of Robot Writers
5 ‘Journobots’ challenging the role of traditional journalists
Robots have the potential to take over a massive proportion of jobs – even mine. Did a robot write this article? Soon, you might not be able to tell. Through advanced data analysis, AI powered robots are steadily taking on the role of traditional writers. However, the use of robot journalism hasn’t been plain sailing. Last year, Facebook made the mistake of sacking its human editors and paid the price when the algorithms began posting fake news stories. Since then, the social network has reinstated human writers as a quality check on automated news. At the moment, bots need to be supervised by traditional editors. . . but in a future world, this could all change. Let’s take a look at which journobots are challenging – or indeed enhancing – the role of traditional journalists today.
1. Xiao Nan
In January, Chinese robot journalist Xiao Nan published its first public article. The 300 character piece was debuted in Southern Metropolis Daily, and covered the Spring Festival travel rush. Xiao Nan was developed by an R&D team at Peking University. According to Professor Wan Xiaojun, the bot can also write short stories and longer reports. The 300 character article was created in just one second, which has obvious positive implications when publishing breaking news and time sensitive stories. Like other robot journalists, Xiao Nan uses advanced data analysis to formulate reports.
The Wordsmith platform is owned by Automated Insights and became publicly available in 2015. Users simply input data, and the bot creates a story using ‘branching paths’ that add words, phrases and sections. Since 2006, the Associated Press has used Wordsmith in a variety of different articles. At the moment, the news agency has tasked the platform with covering Minor League Baseball (MLB). By using data from MLB Advanced Media, Wordsmith can now report on 142 teams across 13 minor leagues. Wordsmith can also produce client reports, product descriptions and even financial summaries. . . so perhaps journalists aren’t the only ones who need to be concerned.
One of the earliest examples of data-fuelled robot journalism is Quakebot, a platform which helped the Los Angeles Times to publish an article about the 2014 LA earthquake within three minutes of it happening. However, in 2015 the bot reported a California earthquake with a magnitude of 5.1, based on data from an earthquake that actually happened in Alaska. The information was clearly published too quickly, without time for an in depth quality check. This also begs the question of how far we can trust superfast robot journalism. To make sure that readers know what they’re reading, each piece that’s co-created by Quakebot is published with the tagline ‘this post was created by an algorithm written by the author’.
During the Rio Olympics, a journobot called Xiaomingbot published a total of 450 stories in 15 days, each ranging from 100 to 800 words. The bot created the pieces for Chinese news app and co-creator Toutiao just two minutes after the different events had ended. The actual language used by the bot was described as, well, robotic, but the sheer speed and volume made the platform particularly impressive. Xiaomingbot is another example of robot journalists reporting on sport, which is a useful application for algorithms as it involves lots of information and numbers that need condensing into coherent summaries.
In 2016, Heliograph covered the Iowa congressional district seat election by analysing data from VoteSmart.org as well as recognising wider electoral trends. The article it produced for the Washington Post has been described as the most sophisticated use of AI journalism yet. As well as creating material itself, Heliograph generates tips by flagging up data anomalies and notifying human reporters. The algorithm is given a narrative template with a list of key phrases, and then uses it alongside data input to form a story. Looking forward, Heliograph will be able to automatically update stories as and when additional facts are discovered, enabling fluid journalism.
As brilliant as journobots may be, they aren’t ready to kill off traditional journalism yet. For a start, there are various things that a robot journalist simply can’t do. For example, they can’t currently conduct interviews, ask leading questions or come up with a news angle – yet. Like humans, they can also be wrong (as in the 2015 California earthquake case). Even so, bots are the perfect answer to sorting through reams of data. In the 2012 election, for example, it took 25 hours for four employees to compile enough data to publish a fraction of the results. In contrast, Heliograph created over 500 articles about the 2016 results without human intervention. Nevertheless, if journalism is to retain its quality, algorithms and human reporters need to collaborate to deliver fully informed and reliable material.