By Jason Crawforth, CEO and founder of SWEAR
Originally published to deloitte.com on August 7, 2024
In an era of rapidly advancing artificial intelligence, the ability to create hyperrealistic counterfeit videos, audio, and images, known as deepfakes, poses a significant threat to our society. The implications of deepfakes are far-reaching and not yet fully understood. From attempts to influence elections, as seen in recent elections in New Hampshire1 and Slovakia,2 to the sheer volume of content uploaded every minute, detecting fakes amid this deluge is an enormously complex challenge.
As the AI technology underpinning deepfake models continues to improve, we expect artificial content to be a worsening problem. Seven years ago, creating a deepfake impersonating someone’s voice required around 30 minutes of audio of that person talking to train the model.3 Today, only three seconds of audio is needed.4 Similarly, older video deepfakes couldn’t replicate details like how a person looks while blinking. These problems have been solved, and today’s models can even recreate blood flow patterns in a person’s face.
Read the full article at Deloitte.com
