In addition to leaning on the software to generate minor league and college game stories, The A.P., like Bloomberg, has used it to beef up its coverage of company earnings reports. Since joining forces with Automated Insights, The A.P. has gone from producing 300 articles on earnings reports per quarter to 3,700.
The Post has an in-house robot reporter called Heliograf, which demonstrated its usefulness with its coverage of the 2016 Summer Olympic Games and the 2016 elections. Last year, thanks to Heliograf, The Post won in the category of Excellence in Use of Bots at the annual Global Biggies Awards, which recognize accomplishments in the use of big data and artificial intelligence. (As if to make journalists jittery, the Biggies ceremony took place at Columbia University’s Pulitzer Hall.)
Jeremy Gilbert, the director of strategic initiatives at The Post, said the company also used A.I. to promote articles with a local orientation in topics like political races to readers in specific regions — a practice known as geo-targeting.
“When you start to talk about mass media, with national or international reach, you run the risk of losing the interest of readers who are interested in stories on their smaller communities,” Mr. Gilbert said. “So we asked, ‘How can we scale our expertise?’”
The A.P., The Post and Bloomberg have also set up internal alerts to signal anomalous bits of data. Reporters who see the alert can then determine if there is a bigger story to be written by a human being. During the Olympics, for instance, The Post set up alerts on Slack, the workplace messaging system, to inform editors if a result was 10 percent above or below an Olympic world record.
A.I. journalism is not as simple as a shiny robot banging out copy. A lot of work goes into the front end, with editors and writers meticulously crafting several versions of a story, complete with text for different outcomes. Once the data is in — for a weather event, a baseball game or an earnings report — the system can create an article.
But machine-generated stories are not infallible. For an earnings report article, for instance, software systems may meet their match in companies that cleverly choose figures in an effort to garner a more favorable portrayal than the numbers warrant. At Bloomberg, reporters and editors try to prepare Cyborg so that it will not be spun by such tactics.