Before Jeff Bezos—the man who founded Amazon.com in
1994—purchased the Post in 2013, news content that had been generated by AI was
coming from just a few, small, companies. At the time, the content those AI
tools could produce was number-heavy, and mostly was only used in the stock
market, sports sections, and other applications where statistical analysis was
most important. However, Bezos his employees at The Washington Post envisioned
an AI tool that could generate more editorial content that was well written, well-sourced,
and interesting for news consumers.
According to an article written by Joe Keohane of
Wired Magazine, the “Heliograf” was introduced by The Washington Post in 2016
and was first used to auto-publish stories during the Summer Olympics in Rio.
In late 2016, a more sophisticated version was used to create content about the
upcoming election. According to Keohane, the process involves editors forming a
basic template or story board for the articles they are hoping to generate.
Editors and reporters can then input key phrases and plug the AI system into
sources of organized data. From there, the program isolates significant data,
matches it with the key phrases used, fuses the sourcing (the data) with the
key phrases (the prose), and then generates and distributes distinctive
versions on The Washington Post’s various platforms. According to Keohane the
system can also discover statistical anomalies in the data sets it analyzes,
which in the case of political reporting, can be a great way to get a tip about
changes in polling, for instance.
Essentially, news organizations have two primary
goals when it comes to their AI tools. The first is to use AI in order to grow
their audience. Secondly, news companies hope that AI tools will not ultimately
replace reporters, but rather make their newsrooms more efficient by allowing
their reporters to spend time on stories that AI can’t tackle. Traditionally,
news organizations had to target a relatively large audience with a relatively
small number of time-intensive stories written by human beings. With an AI
system that can create compelling content, organizations can target multiple
smaller-sized audiences with a massive number of articles written by AI about
niche subject matter. Larger news organizations are also using AI tools to
generate more localized content. With many local news organizations struggling,
larger organizations are hoping to use their AI systems to generate the local
content that consumers are increasingly without.
The Washington Post, like most every news company, is
hoping to genereate new revenue streams, and is supposedly in discussions to
authorize use of its Content Management System to clients like Tronc, a
conglomerate that includes the L.A. Times, the Chicago Tribune, and several of
other regional and local papers. As companies continue to struggle with the
decreasing amount of resources, it’s not hard to imagine a future in which AI
plays a larger and larger role in creating journalism.
Several other news outlets have some form of
artificial intelligence. For example, “Wibbitz” is an automated program used by
USA Today in order to produce short media content on the fly. It has the
ability to summarize and shorten news articles into a short script, storyboard
and put together a piece made of still images and raw video footage, and can
even add robotic narration to the video. Reuters has a tool that they use to
help measure the reliability of posts on the popular social media site Twitter.
An algorithm ranks developing stories based on how much news value they have,
how accurate they may be, how popular they are on the platform, and check what
others are saying about a particular news event. The popular website BuzzFeed
originally developed their AI tool “BuzzBot” to crowdsource reporting from the
Republican and Democratic National Conventions, however, BuzzFeed decided to
open-source their software, making “BuzzBot” the foundation for many reporting
tools that are currently being developed with an AI component.