The fake photos of the celebrities and world leaders have been hitting the internet for long. But the new threat of doctored and fake images concerns the pictures of the Earth. And you are wrong if you think the issue is trivial. The image-fakery of a territory must worry us the most. If you ask why, then read on.
As the world comes to rely more and more on the open-source images to understand the physical terrain, just a little of the manipulated data into the supply line of open-source images could create havoc. And imagine the extent when someone would do that purposefully? And China is said to be doing just that.
As per the information shared by the Todd Myers, automation lead for the CIO-Technology Directorate at the National Geospatial-Intelligence Agency, China is using an emerging technique – called generative adversarial networks (GANs) – that may give a fake image of a specified territory.
GANs can be used to trick the computers to see such objects (like bridge or river) in the satellite image of a landscape that aren’t there.
Myers explained that the computer-assisted imagery may make defense-mission planners gather that a bridge crosses an important river. The tactical perspective or mission-planning will train the forces to go a certain route, toward the bridge, but the bridge will not be there. And that will certainly dent the planned mission of the adversary.
GANs was first described in 2014. It represents a big evolution in the way neural networks would see and recognize objects. It may even detect truth from fiction.
It is the conventional neural network that helps in figuring out what objects are what in satellite photos.
The network breaks the image into multiple pieces, or pixel clusters, calculates how the broken pieces relate to one another, and makes a determination about what the final product is, or, whether the photos are real or doctored. It is based on the experience of looking at lots of satellite photos.
GANs is better than a conventional neural network as it reverses the process by pitting two networks against one another. For example, a conventional network might say, presence of x, y, and z in these pixel clusters means this is a picture of a cat. But a GAN network might say, “This is a picture of a cat, so x, y, and z must be present. Secondly, What are x, y, and z and how do they relate?”
A lot of scholars use GANs for spotting objects and sorting valid images from fake ones. In 2017, Chinese scholars used GANs to identify roads, bridges, and other features in satellite photos. However, the concern is that the same technique that can discern real bridges from fake ones can also help create fake bridges that even AI can’t spot from the real thing.
Can military defeat GAN?
The military and intelligence community can defeat GAN. But it is a time-consuming and a costly exercise. It would require multiple, duplicate collections of satellite images and other pieces of corroborating evidence.
The assessing authority needs to have a duplicate copy from different sources. One just cannot trust just one source. This poses both technical and financial challenges.
In addition to the security reasons, the bonafide institutions from the whole world agree that data integrity is a rising concern.
Protecting the open-source data and images is difficult as it is used by everybody – form news organizations to citizens to human rights groups to hedge funds – to make decisions.
Here, the “truth” that the government can access and the “truth” that the public can access is demanded to be the same but that demand is eroding the public credibility of the national security community. The functioning of democratic institutions is also in question due to data sharing.
Experts say that there is an existential battle for truth in the digital domain. The help of the private sector and data providers is inevitable to protect data integrity.