Google’s Jigsaw has a new tool that will help journalists flag doctored images

Jigsaw has unveiled a tool that can help journalists spot fake images on the Internet. Called the Assembler, Jigsaw (a company owned by Google's parent company Alphabet) worked with Google Research and academics to create this platform.

According to Jared Cohen, the founder and CEO of Jigsaw, the work on Assembler began in 2016.

"Together with Google Research and academic partners, we developed an experimental platform called Assembler to test how technology can help fact-checkers and journalists identify and analyse manipulated media," Cohen wrote in a blog. Jigsaw also announced Assembler on its research publication called The Current.

Cohen called Assembler an "early stage experimental platform" and that means that it is not available to everyone yet. Jigsaw has worked with various fact-checking platforms including Agence France-Presse, Animal Politico, Code for Africa, Les Décodeurs du Monde and Rappler to test how Assembler can be used by journalists and in newsrooms.

The blog post also explains that while Assembler is presented as a single tool, it actually is an amalgamation of various "image manipulation detectors", each of which can detect specific types of manipulations.

"Experts from the University of Maryland, University of Federico II of Naples, and the University of California, Berkeley each contributed detection models," Cohen explained.

Assembler uses all these tools to flag and inform journalists and fact-checkers about the possibility of an image being fake or manipulated. The platform also has a system for detecting deepfakes built inside. In all, Assembler has seven detectors that work on an image and look for "specific types of manipulation, from photoshop to general adversarial networks (GANs) which are used in creating deepfakes".

Jigsaw has also built a model that can take signals from all the other detectors inside it and "analyse an image for multiple types of manipulation simultaneously". Cohen claimed that this makes this last model (that has all the tools together) more accurate than the individual detectors.