Researchers can now use AI and a photo to make fake videos of anyone

78983

Researchers have recently developed a way to create fake videos of any person by using a single image with a trained system based on artificial intelligence (AI). This new technology has people worried, particularly in relation to the 2020 presidential election. It is expected that this new AI system will be used to create fake videos of presidential candidates. This may pose a problem to not only the next presidential election, but future elections as well.

The Samsung AI center and the Skolkovo Institute of Science and Technology, both based in Moscow, revealed the development this week. One main point was that the new AI system has an outstanding ability to learn about important aspects of a person’s face. Through this recognition, the AI system can develop a realistic looking video of a person using only their face.

A video was posted this week on YouTube as well revealing how convincing the system can be. The faces used in the video included the famous physicist Albert Einstein, actress Marilyn Monroe, and surrealist painter Salvador Dali.

Advertisement

Despite the realism of the videos, key details of the face for each person were missing or out of synch. For example, the famous mole on the cheek of Monroe was absent from the videos.

The new AI system is similar to deepfakes, which makes realistic fake videos and audio by using impressive AI technology as well. Deepfakes is a combination of the term deep learning and fake. Both systems use a machine learning method called generative adversarial networks (GANs). Deepfakes is different than the new AI system in that they are typically generated by using a video of a person along with a video of someone else acting a certain way. This new AI system only needs the face of someone to make an effective manipulated video.

The use of fake videos is a concern for people, particularly individuals in the political and intelligence communities because they are worried the fake videos could be used to sway the opinions of voters. What is worse is that the videos do not need to be altered to be affective. For example, a video of House Speaker Nancy Pelosi went viral this week that was not altered, but rather just slowed down to make it appear as though Pelosi was slurring her speech.

The development of the new AI system is still in the early stages. The new AI system so far has only been trained to create a person’s head, neck, and some of the shoulder.

Siwei Lyu studies deepfakes and is the director of the computer vision and machine learning lab at the University of Albany in New York. Lyu revealed to CNN Business that the new research could make it easier to develop deepfakes with significantly less data than is currently needed. Today, you need at least thirty seconds of video in terms of the person you want to fake and another person who must also be recorded doing movements that you desire in the doctored video. This new AI system will make it even easier than this to create a fake video.

Lyu did say that without the needed data, the quality of the fake video, even with the new AI system, will be limited.

Advertisement