Algorithms, smart robots which sort loads of information and arrange it according to users’ requests, are entering the newsroom. This automated-manual coexistence already occurs when journalists try to make content “facebookable” or Google-friendly. To do so, they rely on algorithms roughly based on the same principle as those used for displaying “most sent/commented/blogged articles”. Robots took over human for such tasks.
What is the next step? Will such robots become standalone content creators working like online news Trojans? A piece from The Week wonders: “Are computers the cheap, new journalists?” Behind these questions, a software program designed by Narrative Science, a Chicago-based startup, can write articles made up of … comprehensible sentences.
How does it work? An algorithm agreggates data and converts it into articles. Until now, this technology has been effective only for sports pieces. But Narrative Science says that this technology can apply for finance (by scanning companies’ financial accounts) and politics (through opinion polls and election results).
Beware, Journalists… Artificial intelligence researcher Kris Hammond, who has been quoted by The New York Times, predicted that “within five years, a computer will win a Pulitzer Prize”. “For very simple, brief information, automated writing can do the trick”, says Frédéric Filloux, editor or the Monday Note and a professor at Sciences Po School of Journalism. “But for everything else, you have to look at the pole vaulting theory: Everyone can jump 5.9 feet (Editor’s note: with at least a bit of practice), but not many can jump 7.5 feet. That’s the difference between very good and great.”
Not human, no journalistic soul
Alexandre Malsch, a 26-year-old engineer who heads Melty Network, agrees with this view. “A robot will never be able to make a play on words, unless you record all existing wordplays in a huge database… In any case, a robot will never be able to write an article with soul.” And Malsch knows a lot about robots. About thirty algorithms permanently scan his Melty.fr website aimed at teenagers (four million unique visitors). Its goal is to “help” writers produce content in the right format, on the right subject, and at the right time. (Needless to say, there is no point in posting a story about Lady Gaga when the target audience is at school.)
Heading towards all-automated journalism?
In order to be highly visible on internet search engines, the young developer designed a publishing tool called CMS (Content Management System) that has been providing journalists since 2008 with a “practically all-automated” system. What length does the headline have to be to appear among Google’s top hits? “No writer can calculate the optimal length; only a robot can”, says Malsch.
Indeed, in CMS, the robot turns a title to green when it has the right length and to red when it is too long or too short. Same goes for key words used in a title. The writer can suggest three different titles for each content, and the robot gives each one a success rating, so the writer just has to select the title which gets 90 or 95%. In Melty’s CMS, another parameter outsourced to robots is the number of links that a specific article should contain. Robots go further than that: they also say at what time the article should appear online by analyzing a huge amount of data to figure out when the article will be the most visible on Google – something that is totally impossible for a journalist.
This hit-boosting tool is just one example of the automation being applied to publishing systems. It does not touch or change the text (except for the links and titles). “It’s just like a new Word—a tool that helps journalists but doesn’t replace their jobs, interviews, or analysis.”
Putting a human touch into the work of machines
Is this all impressive? Absolutely. Is there any cause for concern? Maybe, but all-automated journalism is certainly not around the corner. “When you see how difficult it is for translation tools to produce relevant results in real time, you realize that it is not going to happen tomorrow,” explains Filloux. All the more so since before a journalist actually starts to write a piece, he or she must collect source material that is far more important than what he or she finally uses… The original volume should be 5-10 times greater than the actual published content.”
As a (paradoxical) result, Malsch and his development team are injecting manual work back into the machines, especially for editing and selecting content, and for forcing the publication of content in real time rather than waiting for the robot to get it out. “As the world is moving forward, human choice becomes increasingly important,” concludes Malsch. “Manual work is regaining currency.”
Note: This article was written by a human being.
Alice Antheaume (translated by Micha Cziffra)lire le billet