patrick.net

What are Deepfakes?

A combination of the terms “deep learning” and “fake”, deepfakes are synthetic media created using deep learning technology. Synthesizing images, video, or audio, may have no obvious nefarious purposes. However, manipulating media using the images, videos or voices of real people pose moral and legal concerns.

Deepfakes can portray people performing actions they never did or saying things they never said. Basically, by feeding the models hundreds to thousands of target images, deepfake algorithms learn what someone’s face looks like from multiple angles and in a variety of expressions. Through training, the algorithm can then predict what the target individual’s face would look like mimicking the facial expressions of someone else. A similar process is used for training a deepfake algorithm to mimic the accent, intonation, and tone of someone’s voice.

Furthermore, the technology requirements to run deepfake algorithms are low. Any motivated individual with a mid to high-grade consumer PC and enough storage space can create deepfakes. To learn more about deepfake algorithms, check out the open source programs DeepFaceLab and FaceSwap which are both available on Github. ...

While many people are familiar with photoshop and the ability to alter images, deepfake technology, especially voice mimicry, is unknown to almost everyone outside of the data science and machine learning fields. Manipulated audio or synthesized speech can be used in virtual kidnapping scams where victims are often targeted through social media and receive phone calls demanding ransom for a loved one. ...

The problem, Watts explained, is that as media technology grows, audiences are more deeply engaged. Coupled with high engagement, it is incredibly difficult to refute a falsified video after it has been watched. The two clearest dangers Watts emphasized are the targeting of U.S. officials and democratic processes, and the incitement of physical mobilization under false pretenses.