Jigsaw, a technology incubator at Google, has released an experimental platform called Assembler to help journalists and front-line fact-checkers quickly verify images.
How it works: Assembler combines several existing techniques in academia for detecting common manipulation techniques, including changing image brightness and pasting copied pixels elsewhere to cover up something while retaining the same visual texture. It also includes a detector that spots deepfakes of the type created using StyleGAN, an algorithm that can generate realistic imaginary faces. These detection techniques feed into a master model that tells users how likely it is that an image has been manipulated.
Why it matters: Fake images are among the harder things to verify, especially with the rise of manipulation by artificial intelligence. The window of opportunity for journalists and fact-checkers to react is also rapidly shrinking, as disinformation spreads at speed and scale.
Not a panacea: Assembler is a good step in fighting manipulated media—but it doesn’t cover many other existing manipulation techniques, including those used for video, which the team will need to add and update as the ecosystem keeps evolving. It also still exists as a separate platform from the channels where doctored images are usually distributed. Experts have recommended that tech giants like Facebook and Google incorporate these types of detection features directly into their platforms. That way such checks can be performed in close to real time as photos and videos are uploaded and shared.
There are other approaches to consider, too. Some startups are pursuing verification technology, for example, which memorizes the positions of pixels in a photo at the moment of their capture—but this also comes with challenges.
Beyond technology: Ultimately, technical fixes won’t be enough. One of the trickiest aspects of digital fakery isn’t the fake images themselves. Rather, it’s the idea that they exist, which can easily be invoked to doubt the veracity of real media. This is the type of challenge that will require social and policy solutions as well.
Why Meta’s latest large language model survived only three days online
Galactica was supposed to help scientists. Instead, it mindlessly spat out biased and incorrect nonsense.
DeepMind’s game-playing AI has beaten a 50-year-old record in computer science
The new version of AlphaZero discovered a faster way to do matrix multiplication, a core problem in computing that affects thousands of everyday computer tasks.
A bot that watched 70,000 hours of Minecraft could unlock AI’s next big thing
Online videos are a vast and untapped source of training data—and OpenAI says it has a new way to use it.
Google’s new AI can hear a snippet of song—and then keep on playing
The technique, called AudioLM, generates naturalistic sounds without the need for human annotation.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.