advertisement
advertisement

‘It’s a prodigious problem’: How Google’s Jigsaw is sniffing out faked images

The Google think tank is releasing a bundle of visual disinformation detectors, a platform where researchers can collaborate, and a new publication to educate the public.

‘It’s a prodigious problem’: How Google’s Jigsaw is sniffing out faked images
[Photos: DESIGNECOLOGIST/Unsplash; Jack Millard/Unsplash]
advertisement
advertisement

As the threat of digital misinformation looms large before the 2020 elections, Google’s Jigsaw research group is jumping in to help sniff out attempts to alter images for political purposes. The group is releasing a bundle of visual disinformation detectors to help fact-checkers and journalists identify images that have been manipulated. These tools will live on a platform that doubles as a place for researchers to collaborate on different detection techniques. It’s also announcing a “research publication” called The Current, where people can access information about disinformation and Jigsaw’s and its partners’ efforts to fight it.

advertisement
advertisement

“We want to be working with the people who are on the front lines of countering disinformation, and we’re really trying to look around the next corner so that we can build and ship tools that can address future manifestations of the problem,” says Jigsaw’s CEO Jared Cohen. “It’s a prodigious problem.”

Jigsaw was originally established in 2010 as Google Ideas by Eric Schmidt, who described it as a “think/do tank” to research issues at the intersection of technology and geopolitics. But the incubator hasn’t talked about its disinformation work until now, Cohen says.

For more than a year, Jigsaw has been quietly working on artificial intelligence approaches to help counter the threat of deepfakes and other forms of image manipulation. Today it’s presenting the main thrust of its work, an open research platform called Assembler containing a set of experimental detection tools that journalists and fact-checkers can use to analyze manipulated media, as Cohen explained in a blog post yesterday.

Developed by Jigsaw (with input from Google Research) and its partners in academia, the tools use computer vision AI to show the probability of specific types of manipulation for a single image. One tool might recognize that somebody has likely messed with the brightness of an image to create a certain look. Another might see signs that something has been copy and pasted into an image to change its message.

Jigsaw developed the tool in Assembler for detecting deepfakes—images or video manipulated with AI to portray people doing or saying fictional things, sometimes for political reasons. The tool, called StyleGAN, uses two separate neural networks working against each other to identify image traits with greater and greater accuracy. Another tool, which Jigsaw calls its “ensemble model,” is trained with data from each of the individual image scanners on the platform, allowing it to analyze an image for multiple types of manipulation simultaneously.

Jigsaw’s Disinformation Data Visualizer analyzes research from the Atlantic Council’s DFRLab on “select coordinated disinformation campaigns.” [Animation: Jigsaw]

Experts from the University of Maryland, University Federico II of Naples, and the University of California, Berkeley each contributed detection models, and the platform also provides a space where researchers of detection technologies can collaborate with one another.

advertisement

A handful of newsrooms around the world, including Agence France-Presse, Mexican publication Animal Politico, and Code for Africa, are now testing the Assembler tools, with their feedback looping back into the development process.

“For journalists, we want to be able to provide an additional set of signals that might help them tell if the images they’re seeing are fact or fiction,” Cohen says.

Alongside its tool development work, Jigsaw also announced a new web publication called The Current, which it will use to showcase its ongoing research on an array of subjects. The first issue is all about disinformation. It lays out research on the architecture of disinformation campaigns, the tactics and technology behind them, and how new tools are being used to detect and stop fake information.

[Animation: Jigsaw]
The Current‘s first edition will also feature a new tool Jigsaw developed called the Disinformation Data Visualizer, which visually maps out the research from the Atlantic Council’s DFRLab on “select coordinated disinformation campaigns around the world” and shows the specific tactics used and countries affected. Jigsaw wanted to share it “to enable a dialogue about the most effective and comprehensive disinformation countermeasures,” Cohen says.

The Current, Cohen told me, will also feature broad categories where Jigsaw can regularly publish its research work in a number of areas.

“Our main motivation was creating a place where people can access the work of many experts and independent researchers and organizations that are on the front lines of dealing with this problem,” Cohen says.

About the author

Fast Company Senior Writer Mark Sullivan covers emerging technology, politics, artificial intelligence, large tech companies, and misinformation. An award-winning San Francisco-based journalist, Sullivan's work has appeared in Wired, Al Jazeera, CNN, ABC News, CNET, and many others.

More