Impressionist and actor Jim Meskimen is known for his voice acting for various Hollywood movies. Last week, he released a video collaborating with a YouTube video creator named “Sham00k”.
Meskimen is skilled in doing accurate voice impressions of famous actors and politicians such as Robert DeNiro, George Clooney, Jack Nicholson, George W. Bush, and more. “Sham00k” is a content creator who uses AI and machine learning technology to make deepfake videos, blending celebrity faces into popular scenes from movies. Together they created a deepfake video in which Meskimen flawlessly imitates both the voice and the face of 20 different actors during the recitation of a poem.
[Related article: Using AI to Combat Deepfakes and Fake News]
Deepfake AI has been prominent since late 2017, when a Reddit user named “deepfakes” would create videos involving celebrity faces being digitally transplanted onto the bodies of celebrities, such as Nicolas Cage’s visage being placed into different movies which he did not star in.
Using pattern and image recognition, computers can easily create videos with faces swapped, keeping the natural framework of the head while swapping out the corresponding facial feature with a substitute image. Applications such as Face Swap or DeepFaceLab can take a reference video and learn the facial structures and variations. The program itself can perform facial recognition to identify the common features such as the eyes, nose, and mouth in order to align the two faces. By feeding the computer with more images and videos of the celebrity’s face and mannerisms, the computer will learn the donor face better, giving the DeepFake a more smooth integration.
As with many of the recent deepfake videos online, “Sham00k” and Meskimen’s video is a fun and entertaining application of machine learning technology. There is a darker side to this technology, however. Deepfakes have also been used to create a video of prominent business leader, Facebook’s Mark Zuckerberg, expressing opinions that are not his own. As machine learning technology progresses and deepfake videos become increasingly difficult to detect, it’s possible, even likely, that they will have very real adverse effects on businesses, the government, and private citizens alike.
[Related article: The Bright Side of Technology in Black Mirror]
The future is not without hope, however. Facebook, in partnership with several companies and academic institutions, has created a challenge to incentivize the development of tools that can help detect deepfakes. And hopefully, as we continue to be entertained by deepfakes like “Sham00k” and Meskimen’s, we’ll become better able to detect them on our own.
Ready to learn more about deepfakes? Check out the ODSC West 2019 talk, “The New Face of Biometrics In The Era of Hacks, Fakes, Bans, and GANs,” by George Wiliams, the Director of Data Science at GSI Technology on Wednesday, October 30, 2019.