You might have seen the morphed and photoshopped picture which has deceived you in one way or the other. To make you confused, even more, a website has been generating realistic face faces. After ThisPersonDoesNotExist.com, a new website named WhichFaceIsReal.com has been introduced to let you distinguish between real and fake faces.
Jevin West and Carl Bergstrom, two academics from the University of Washington introduced the site to aware people about the existence of this technology. Both of them study how information spread in society and has been trying to see how AI-generated fake faces can cause trouble and can create unfaithfulness in evidence.
Bergstrom said that often technology is there but people do not know about it. Their lack of information about the newly intricated AI could affect them greatly in one way or the other. It is important to educate people about how realistic and fake faces can be generated just like people were aware of how distortion image can be created using Photoshop.
To generate these faces, a machine learning method called GAN, Generative Adversarial Network is used. A lot of data, like portraits of real people, is fed into the network. The pattern of these pictures is learned and then replicated by the network.
Screenshot: Which Face is Real
To generate a perfect picture, GANs test themselves. Faces are generated in one part of the network, while the other compares the generated picture with the training data. In case a difference is identified, the image is sent to the drawing board for improvement.
There are certain limitations, but these techniques can be used to manipulate audios, videos, and pictures. It also allows other features like turning videos of people into puppets or show you a dancer.
The brains behind the site said it can be used to spread misinformation in a society. Like a fake image od, a culprit can be spread online, which cannot be tracked by anyone using Google’s Reverse Image search.
It is also expected that in the coming few years, this technology will improve making it even harder to distinguish between the real and fake generated image.
Jevin West and Carl Bergstrom, two academics from the University of Washington introduced the site to aware people about the existence of this technology. Both of them study how information spread in society and has been trying to see how AI-generated fake faces can cause trouble and can create unfaithfulness in evidence.
Bergstrom said that often technology is there but people do not know about it. Their lack of information about the newly intricated AI could affect them greatly in one way or the other. It is important to educate people about how realistic and fake faces can be generated just like people were aware of how distortion image can be created using Photoshop.
To generate these faces, a machine learning method called GAN, Generative Adversarial Network is used. A lot of data, like portraits of real people, is fed into the network. The pattern of these pictures is learned and then replicated by the network.
Screenshot: Which Face is Real
To generate a perfect picture, GANs test themselves. Faces are generated in one part of the network, while the other compares the generated picture with the training data. In case a difference is identified, the image is sent to the drawing board for improvement.
There are certain limitations, but these techniques can be used to manipulate audios, videos, and pictures. It also allows other features like turning videos of people into puppets or show you a dancer.
The brains behind the site said it can be used to spread misinformation in a society. Like a fake image od, a culprit can be spread online, which cannot be tracked by anyone using Google’s Reverse Image search.
Also Read: This New Artificial-Intelligence-Based Text Prediction Tool Can Write Better Than Many HumansAlong with this, many researchers are also working on tools that will be able to identify these deep fades. West thinks it is possible to spot fake images just by noticing the fake hair, unsymmetrical faces, misaligned teeth, and other few little hints.
It is also expected that in the coming few years, this technology will improve making it even harder to distinguish between the real and fake generated image.
