The Associated Press Uses AI To Boost Content And Video Volume

Like most publishers, The Associated Press (AP) is evaluating ways it can increase video and article output without straining its staff or resources.

A few years ago, the AP began researching ways it could leverage artificial intelligence and machine learning to improve its internal processes.

The publisher then formed a cross-functional AI committee to vet new technology and identify tools to address business problems.

“The way we operate is to identify proofs of concept or small experiments that we can then scale,” said Francesco Marconi, AP’s strategy manager and co-lead on artificial intelligence. “Something we’re doing right now is testing different image recognition and computer vision tactics to tag photos.”

It’s a critical capability for the large-scale news agency, whose content is used by other publications and journalists to develop their own localized editorial.

With such a large repository of articles, imagery and video, AP uses metadata to comb through and identify relevant content.

“It’s really important that when a journalist is looking for a certain type of imagery, they can find what they’re looking for,” Marconi said. “Enabling that via computer vision is one way we’re doing that.”

Because most of AP’s revenue comes from licensing content – its advertising operation is relatively small – it’s always looking at ways to unlock value from its content and its archives.

Some of AP’s earliest successes in applying AI to content production use data to expedite news-story production.

The publisher tapped Automated Insights, which uses machine learning and natural-language processing to automate text for sports updates and quarterly earnings stories augmented with real-time trending information from Zacks Investment Research.

“To give you a sense of the impact of this first project, we went from producing about 300 stories to close to 4,000 each quarter, which was a 12x increase in content output,” Marconi said. “We also saw a reduction in error rate and were able to free up 20% more of reporters’ time to focus on higher-value [projects].”

Using machine learning for text automation whet the AP’s appetite for more ways to improve its editorial and commercial processes.

One major focus is video and imagery, given publishers’ hunger to produce more video in formats ranging from live reports to shorter news clips.

The AP is researching ways it can use the text-to-video platform Wibbitz, which helps pubs create companion videos for graphics or list-based articles that need a visual element.

Wibbitz uses natural-language processing and advanced algorithms to automatically generate videos from related story images or infographics.

“We can use these tools to automate things that are a little more repetitive for us to do internally, such as adding graphics automatically or [adding images or video to] profiles for people that take awhile to create,” Marconi said. “Our expertise is in storytelling, and when you’re a large organization, sometimes innovation can move more slowly. So we partner with a lot of startups to find pockets for new ideas.”

The AI committee does not mandate all technology decisions for the organization.

A tool must work across editorial, product and technology teams to be deployed more widely. Marconi insists automation is not replacing the human component.

“We’re getting to a place where we use AI-assisted tools to enhance work that is developed by our human journalists,” he said, “so this idea of human-machine collaboration is very important.”

This post was syndicated from Ad Exchanger.