As full-scale automation looms over employees, poised to turn the economy on it’s head (for better or worse), many wordsmiths consider themselves safe from the threat of robots and AI. After all, there’s nothing like a human hand in writing. Right? Automation is typically seen as applied to manual labor, jobs in service, and systematic tasks like accounting. But could algorithms ever report the news?
Newsflash: They already are!
While some journalists are using robots to attend conferences at a distance (such as in the video below) others are using AI as more than just a stand in for attendance.
Every three months the Associated Press, one of the world’s most trusted news outlets, publishes some 3,000 articles written by an automated system fine-tuned to the AP Style Guide. The articles aren’t in-depth and have a particular focus on the oft dry coverage of business reporting of quarterly earnings. But where AP used to cover the earnings of some 300 companies, they now publish ten times more reports. Automated Insights – the company behind the platform Wordsmith – is credited with the byline.
Wordsmith writes its articles sort of like a game of ad libs. Users create syntactic paths around which information can be uploaded and modified. For example, with the sentence “[Company] has seen an [increase/decrease] in earnings of [#] percent” Wordsmith fills in the brackets with the appropriate information for the relevant company. These paths can be applied to any number of topics. “[Football team] beat [football team] by [#] points” or “[Film] opened this weekend’s box office with $[#]”. These templates generate to cookie cut articles but serve their purpose then what’s needed is quick, easy, and comprehensible data. The system goes deeper though with branches that enable words and phrases to be added and modified conditionally, depending on the words and phrases surrounding them. For example 25MPH may be fast for a human sprinter but not for a motorcycle. By creating branches, users can make certain that Wordsmith uses the right references and relations between ideas.
AP isn’t the only organizations using Automated Insights whose public relations manager James Kotecki told The Verge that Allstate, Comcast, and Yahoo also use Wordsmith’s automated articles to report things like fantasy football stats. The system is also designed to produce technical texts like business reports and commercial texts like product descriptions for companies.
Since Wordsmith’s templates aren’t yet dynamic, the system is mainly used for statistical and superficial reporting – it’s too difficult to cover the nuances of interviews and analysis which are better written by trained journalists. Like trained journalists, the initial system wasn’t without errors. It also read a bit like an algorithm. According to Philana Patterson, an assistant business editor at the AP, some staff members pegged the articles as slightly inhuman with their drab tone and essentials. So when the system began to generate stories in July 2014, every published article was scanned by an editor in order to check for mistakes. This feedback helped Automated Insights refine their system, which now apparently releases fewer errors than their human counterparts.
Still, AP and Automated Insights told The Verge that this automation has lead to zero layoffs. Though Wordsmith can write thousands of basic articles, many of these still need human interaction to review style and contribute reportage too complicated for the system. Rather than becoming redundant, the journalists who used to write those dry quarterly earnings articles have had their time freed up to pursue unique angles, deeper interviews, and more critical analyses.
Beyond statistical accuracy, one of the big selling points for Wordsmith is the system’s speed. Kotecki estimates that at peak performance the technology can create 2,000 articles per second. But an NPR experiment supported that old adage about quality of quantity. NPR’s feature report had Wordsmith challenge the news organization’s own White House correspondent to write and publish the same story. Though Wordsmith outdid the correspondent in speed, NPR listeners voted that the human written article was richer and more engaging.
In October of the year, Automated Insights opened Wordsmith up to the public, enabling any interested party to sign up and toy around with their beta version. General access is scheduled for January This will grants individuals outside the media with hands-on access to a tool that may well change the way media functions.
In contrast to traditional media which seeks to publish limited stories with broad appeal and high caliber, Automated Insights looks to publish many stories with vary specific target audiences. In a press release, Automated Insights CEO Robbie Allen emphasized his company’s goal of personalized content. “Instead of writing one story and hoping a million people read it, Wordsmith can create a million satires targeted at each individual user and their preferences. It’s a story that is totally unique to each user because it is powered by their data.”
The open access system and desire to develop highly personalized stories creates some concern around journalistic integrity. What information will be communicated and to whom? Tom Kent of the AP chimed in early this year with an ethical checklist, eleven questions intended to keep the writing system clean and beneficial for readers. Among the list are calls for disclosure that the stories are automated, human monitoring of the machine, and foresight to prepare for the controversy that will undoubtedly arise as more of our news stories are created by AI.