TechRaking Boston: Verifying the News

Making sure the imagery you see is real and has not been manipulated

By Matt Carroll <@mattatmit>

TechRaking Boston was a conference about verifying imagery. The one-day conference at the MIT Media Lab in Cambridge, MA, drew together about 50 technologists and journalists from around the country and Europe interested in this a fast-growing field. This was the first conference devoted wholly to the topic, and we hope it becomes an annual event.

The surge in interest is driven by the sheer number of newsworthy images — both video and pictures— that are uploaded every day on a number of social media sites. While the vast majority of the millions of videos and pictures show exactly what they purport to show, some have been manipulated or misrepresented.

Verifying images is a huge problem for news organizations, which stake their reputations on the accuracy of their content. Fraudulent imagery is a nightmare for everyone. Readers are deceived and irate when they find out they have been fooled; governments and institutions can make bad decisions based on deceptive imagery; journalists are furious about displaying and writing about bad information.

Bad images can range from the eagle that snatched the baby in the park to changing the insignia on the plane in the Ukraine that was hot down.

The goal of the conference was to help people learn more about detecting the frauds and verifying what’s true. To that end, we brought together some of the top experts in the fields to talk about new emerging trends.

We held talks, a panel discussion, lightning talks, and held a mini-hackathon to prototype ideas for fighting the problem. It was chance for people to get together for a day to talk about what they have discovered and where the field is headed.

As we say on the web page, “we will highlight use cases, share information, discuss technology and look for ways to create more widespread adoption of visual verification in newsrooms.”

A tremendous list of people and companies active in this field attended, including Andy Carvin from Reported.ly, Claire Wardle, Tow Center for Digital Journalism, Jen LaFleur from the Center for Investigative Reporting, and Hany Farid, a computer science professor and expert on photo manipulation from Dartmouth College.

Great list of sponsors as well: CIR, Google News Lab, Future of News at the MIT Media Lab, First Draft, the Knight Foundation, and Bloomberg.

— ++ —

Teams working on projects.

And the winner is…

5 pm: Our distinguished panel of judges has picked a winner:

Suddenly (formerly know as “Holy Sh*t”) took first place. The app would be a Twitter alert system for disasters, created through an iterative system that used a team of volunteer journalists.

Runnerup: Sway, which would pull people’s social IDs into one site.

A shoutout to our judges, who did a wonderful thoughtful job considering the work of the participants: Samaruddin Stewart, Pixel Project (and a co-organizer of the conference, with Kristin Belden of CIR); Craig Silverman, BuzzFeed Canada; Jen LaFleur of CIR, and Claire Wardle, from the Tow Center for Digital Journalism.

— ++ —

Prototype ideas are pitched (and the judges comment)

2 pm: The groups pitched their rough ideas, with judges offering comments.

  • The Reliable Image Project or TRIP. Let’s create comprehensive standards for imagery that can help social media. The idea would be to build a consortium that would hammer out guidelines and engineering standards, to help guide social media. That might include some basic level of verification, or perhaps a mark that indicated an image had been verified. It could also include privacy guidelines.
  • Sway: The idea is to try to pull everyone’s social IDs into one place. The problem is solves is that it can be very difficult to ID people who have identities scattered across the social media landscape. Hopefully there would be an incentive to get people to sign up for this, such as they could point friends to this so they could find them on other platforms. The advantage for journalists is they can use one search for a bunch of platforms. (Question from judges: Will people be interested in doing that, or do they want to keep their IDs separate?)
  • Checkit: How can we make audiences and newsrooms care about verification? Maybe a game. Checkit aims to walk people through the verification process, turning people into digital detectives. There would be various levels of skill building. (Comment: At least one other game won a TechRaking events, but games are hard to make fun.)
  • Bullsh*t Detector: (This group actually had drawings.) Let’s build a Twitter alert system for disasters. It would look for tweets using a range of profanity and emojis. It would work like this: People tweet; relevant data is collected from the firehouse; a database extracts each variable to build tables and libraries; volunteer journalists check the output and finetune. Repeat. This could also be used for other issues, such as monitoring problems with public transportation.

Overheard comment from a judge afterwards: “I made friends with everyone. Now I’m going to break everyone’s hearts.” (I think he was joking.)

Pitch for Bullsh*t Detector

btw: What the judges will look for

Here are the criteria judges will use: Originality/innovativeness; scalability/feasibility; promotes user user engagement; promotes adoption across a wide variety of newsrooms.

The (awesome) lightning talks

10:30 am: Lightning talks:

  • Metadata: Potential and Pitfalls, Kevin Connor of Fourandsix:
  • The Simple Science of Verification, Malachy Browne of Reported.ly,
  • Checkdesk: Technology for Collaborative Verification, Tom Trewinnard of Meedan Labs,
  • Conflict Infrastructure, Madeeha Yasin Merchant, Columbia U,
  • A Tale of Two Images, Samaruddin Stewart, Verified Pixel Project,
  • Automating the Search for the First Upload, Cynthia Fang, MIT,
  • The Lost and Found, Jennifer LaFleur, Center for Investigative Reporting,
  • Turning Citizen Media into Evidence, Christopher Koettl, Amnest International.

Great talk on detecting whether a photo has been manipulated

9:30 am: Hany Farid gave a fascinating talk about forensic techniques for discovering whether a photo has been manipulated.

Some takeaways:

  • Each camera has certain characteristics engineered in how it handles pictures that can be picked up by forensics.
  • Changing the size of a certain part of a picture — say a fisherman makes his caught fish a little bigger — leaves telltale changes at the pixel level that can be seen.
  • Shadows are a great technique for determining if something is fake. Used to show the famed video of an eagle snatching a baby in a park was a fake — the shadows of the baby and eagle showed they had been dropped into another video. That’s partly because it turns out that people are very bad at faking shadows in images.
  • If it’s a spectacular shark photo, it’s almost guaranteed to be fake. ;-)

8:30 am, Sept. 12, 2015: We’re starting.

As our web site says: “With the the cell phone camera’s rise in ubiquity, more people are taking newsworthy photos and videos and sharing them directly with social networks and news media.

“While helpful in providing valuable information and context to news events at times, these images can also be manipulated to become misrepresented visuals intended to deceive or sway opinion.

“Designed for journalists, technologists, academics and imaging experts, at TechRaking: Verifying the News, we will highlight use cases, share information, discuss technology and look for ways to create more widespread adoption of visual verification in newsrooms.”

We’ll be tweeting all day long at #techraking, so stay tuned.

We have an eventful day planned, starting with:

  • talks from LaToya Drake, who does media outreach for the News Lab at Google, and a talk on photo forensics from Hany Farid, a computer science prof at Dartmouth College.
  • a panel discussion on “Verifying Under pressure: Understanding the Tools and Techniques of Social Media Verification.” Panelists are Andy Carvin of Reported.ly, Eliza Mackintosh of Storyful, Aric Toler of Bellingcat, and moderator Claire Wardle of the Tow Center for Digital Journalism.

Matt Carroll runs the Future of News initiative at the MIT Media Lab. He can be followed @MattatMIT.

— 30 —