Sarosh: The act of manipulating content photographs has dated back to 1860s, just a few decades after the first photographs were taken. Deep fakes is just leveraging the latest technology and machine learning tools to automate swapping out faces in video, swapping out voices and audio, and just making it really easy to make changes to content.
Elie Khoury: Yeah, so basically it’s relying on very, very sophisticated machine learning tools that are deep now network. This technology benefits a lot from the increased computational efficiency that happened in the years, including the use of GPUs. Add to this the increased amount of additional data that is available on social media. I mean, it’s easy, for example, to find hours and hours of President Trump to go later and use this data together with those machine learning techniques to create a very high-quality voice of President Trump.
Sarosh: So this was a Reddit user by the name of Deep Fake who was going frame by frame in adult content and swapping out actresses with celebrities. I think that’s where it really originated from, where you could tell that somebody was taking content and swapping out that content. And it wasn’t even using deep neural networks at the time, I don’t believe. I think it was just a basic tool to change faces frame by frame. But I think it sort of escalated quickly after that using machine learning tools and the name stuck around.