Who Would You Believe a Journalist or a Machine?
January 18, 2016 • By Atlantic Re:Think • Blog Post
IN THIS ARTICLE
- Newsrooms are using technology to find, report and tell stories
- Companies like Banjo, CrowdTangle, Spike and Dataminr provide new data-driven ways to break news
Tech companies are helping the world's newsrooms gather and tell stories with advanced machine-learning systems that can scour the depths of the Internet
The journalist made an offhand observation, and, in a minute, he had a new understanding of the world.
He and his colleagues were standing inside what California-based Banjo calls its “24/7 War Room” in Las Vegas, waiting for a demonstration of how the company’s technology can give newsrooms new ways to find, report and tell stories. He happened to mention that the weather outside seemed unusually cold for Las Vegas, maybe cold enough to make people switch from iced to hot coffee.
“I told him to ask the crystal ball to show every single person drinking something at Starbucks right now,” Banjo CEO, Damien Patton, says.
The journalist looked at Patton, incredulous, but he did as he was told, then watched the screens, not for minutes but mere seconds.
“Our system picked up the Starbucks logo on the coffee cups and showed all these photos all at once,” Patton says. “They were blown away.”
Achieving a God’s-eye view of everything happening everywhere right now may sound like something out of The Matrix, but that’s exactly what companies like Banjo are trying to do. They take huge feeds of Big Data from countless digital sources—everything from the 500 million tweets uploaded daily to video feeds from Brazil to traffic reports in Albuquerque—and sort through it all live, as it happens online, providing an instant source of breaking news as well as tips, insights and people on the ground long before a reporter could reach the scene.
Banjo is just one tech company racing to build deep-learning systems with neural networks that can recognize not only text but also images, video and more. Given their need for speed, newsrooms are a prime target for anything that makes a potential source out of every person on the planet.
The journalist can search for what she wants, as in the case of Starbucks cups or can preset the system to send her email, text messages or other alerts. Learning about all kinds of events, from a small house fire in Pittsburgh to a mass shooting in Paris, becomes as simple as receiving a smartphone notification. If you’re an assignment editor at a TV station and hear something on the police scanner happening at 123 Main Street, you can type in the address and see not only every social media post being uploaded from that location but also how to contact the people doing the uploading, long before a traditional video news team could be there.
Uploaded messages, pictures and video are already being mined to cover places too dangerous to send journalists—the war in Syria and ISIS-occupied Iraq, for example. But instant, real-time access to every hotspot in the world allows for news coverage that is both global and hyper-local.
Banjo is just one company at the forefront of this emerging field. There’s also CrowdTangle, which conducts “social listening” to find trending Facebook posts. A system called Spike, offered by News Whip, synthesizes Twitter and Facebook simultaneously.
Perhaps the most stunning example of big news broken by advanced data analytics came from New York-based Dataminr, which—with just 19 tracked tweets—alerted its users that Osama bin Laden was dead 23 minutes before the first major news network was able to confirm it.
“In the past, journalists had to wait until something had been retweeted by someone they follow or reported by a news wire to learn about it,” says Steven Schwartz, President of Dataminr’s News Division. “With Dataminr, firsthand accounts of breaking news events are delivered directly to them in real time. In some cases, that kind of relevant information can come from sources who have only a small number of followers on Twitter and are therefore unlikely to be followed by the world’s media, yet who happen to be witnessing a major event unfolding in front of them.”
At Banjo, Patten calls those kinds of uploads “event content.” Recognizing that type of content is what makes these systems so powerful. While the web is constantly flooded with cat videos and photos of what our friends are eating for lunch, it’s much rarer for content spikes to occur in unusual places—say, at the scene of a cruise-ship disease outbreak.
“When there’s an unplanned event now, sometimes even before people call 911, they pick up their phones and start recording,” Patten says. “Because of that, event content far exceeds the norm of what people hear about on a day-to-day basis.”
For example, systems like Banjo, after seeing a spike in video uploads from the scene of a tornado, not only verify the event by comparing the upload to an earlier image on Google Earth but also tell the journalist how to reach the person uploading the pictures.
“From the moment someone takes a photograph or a video, like with the Amtrak train crash in May 2015, from the moment that’s posted publicly, our system picks it up, looks at it, alerts our team and makes sure it is going out to, let’s say, NBC Philadelphia in that case,” Patten says. “From the moment that photo was taken to the moment it goes to NBC is 20 seconds. We know who took the photo, where they normally are, what photos they’ve taken in the past, does it match up with the scenery that is usually in that place—we have thousands of things our system does to validate that scene. Everything is done in under a minute.”
And these systems can be used globally, locally or for issue-specific reporting. Thanks to Dataminr, India’s Hindustan Times was able to post images faster than many major news organizations after the Taliban bombed Afghanistan’s parliament. The New York Post reportedly uses News Whip’s Spike service to keep tabs on events in the Big Apple. Even the American Kennel Club is said to be using a Big Data news service, searching for stories that have to do with, say, Golden Retrievers.
“There are huge new organizations that don’t even launch their news helicopters anymore unless they check Banjo first to see if they can justify the cost,” Patten says.
And while such comprehensive data analytics technology may seem revolutionary to journalists using these tools for the first time, it’s still in the early stage. New social media platforms continue to emerge. Satellites launch every day that will add more information, and drones will soon be uploading live video feeds from virtually everywhere. And all that new data will require stronger and more sophisticated analytical tools to sort and understand it.
“We’re using a lot of deep learning, and the field of deep learning is still fairly young,” Patten says. “Because of that, and because of how fast technology is advancing, this field is very nascent. As the Internet of Things and data continue to grow around us every day, you need a system that can organize it. We as human beings just can’t keep up with it, organize it and customize it for the things that matter to us.”
To Patten and his colleagues in this burgeoning field, the knowledge of everything happening everywhere right now just seems like manifest destiny. “Our phones today can alert us if somebody walks into our house,” says Patten. “Why shouldn’t they alert us when something’s happening anywhere else in the world?”