In this episode, we shift our focus exclusively to AI, and we bring three stories that are centered around AI called Generative Adversarial Networks.
Story #1: Malaria Must Die's global campaign featuring a deep fake of David Beckham eclipsed 400 million impressions globally. The campaign features international soccer star, David Beckham in a video where malaria survivors speak through him in different languages. The video uses sophisticated, AI-driven software that is trained on the subject, in this case, Beckham, and then automatically generates a video of the subject speaking the words originally spoken by malaria survivors in nine different languages. Visually, the faked speech is virtually impossible to detect - it looks like David Beckham actually spoke the words. A behind-the-scenes video explains how the video was made. The startup behind the deep fake video production is Synthesia, and their goal is to apply AI tech to replace today's unscaleable, manual, shoot-and-edit video production with their automated computer generation process that produces an equivalent completely in the lab. TechCrunch
Story #2: Several states are developing legislation to regulate deep fake videos. Deepfakes, in this context, are videos altered using AI that make people appear to say things they actually did not. The AI algorithms are trained using existing video of the subject - they observe and record movement patterns in a subject's face. They then simulate them to make the subject do or say something new. There is a significant commercial and political risk from using this technology to manipulate public perception of political leaders, especially during elections, and of business leaders. Three states, California, Texas, and Massachusetts are preparing bills to tackle deep fake videos, with the most prominent, Texas' SB 751, criminalizes the act of creating a deep fake video "to injure a candidate or influence the result of an election." Government Technology
Story #3: The ALS Association is working with Lyrebird, a company that creates digital voices that mimic real people is cloning the voices of ALS patients so they can continue communicating even after the disease takes away their voice. Called Project Revoice at the ALS Association, it is meant to counter the progressive neurodegenerative disease that often leads to a loss of the person’s ability to speak. Working from a few hours of high-quality voice recordings, the AI technology is able to synthesize and fully recreate the unique essence of the human’s voice and build what they call “a complete digital voice clone.” The technology is the same deepfake, Generative Adversarial Network based AI. The ALS patient then uses an adaptive input device - a kind of modified keyboard - to input what they want to say. The Lyrebird software then speaks the words in the patient’s digital voice.