I’d like to thank the Technical College System of Georgia for inviting me to deliver the talk “Who’s Afraid of Deepfakes” at their 2019 Leadership conference in Savannah, GA. It is a great pleasure to be able to deliver this material to people who are not already familiar with it. It was also a privilege to speak after Jason Poovey of Georgia Tech, who provided a fascinating overview of the current state of AI research and his vision for bringing business, math and technology together.
I also want to thank Adie Shimandle, Billie Izard and Steven Ferguson for organizing the talks. Thank you Elizabeth Strickler, Director, Media Entrepreneurship and Innovation at GSU, for introducing me to Adie.
Thank you to everyone who attended my Deepfakes and Ethics talk yesterday at the AI and the Creative Industries Friday Series. Thanks also to Dr. Adam Spring for driving downtown with me and taking a walking tour of some of the parking lot there.
The talk was filmed and should go up in a few weeks.
The Zao app, by Changsha Shenduronghe Network Technology Co Ltd, was released on the Chinese iTunes store a week ago and was popularized in a tweet by Allan Xia.
It is not currently available through iTunes in the U.S. but with a bit of hard work I was finally able to install a copy. I was concerned that the capabilities of the app might be exaggerated but it actually exceeded my expectations. As a novelty app, it is fascinating. As an indicator of the current state and future of deepfakes, it is a moment of titanic proportions.
As of a year ago, when the machine learning tool Fake App was released, a decent deepfake took tens of hours and some fairly powerful hardware to generate. The idea of being able to create one in less than 30 seconds on a standard smartphone seemed a remote possibility at the time. Even impossible.
The Zao app also does some nice things I’ve never gotten to work well with deepfakes/faceswap or deepfacelab – for instance like handling facial hair.
… or even no hair. (This is also a freaky way to see what you’ll look like in 15-20 years.)
What is particularly striking is the way it handles movement and multiple face angles as with this scene from Trainspotting and a young Obi Wan Kenobi. In the very first scene, it even skips over several faces and just automatically targets the particular one you specify. (In other snippets that include multiple characters, the Zao app allows you to choose which face you want to swap out.)
All this indicates that the underlying algos are quite different from the autoencoder based ones from last year. I have some ideas about how they have managed to generate deepfakes so quickly and with a much smaller set of data.
Back in the day, deepfakes required a sample of 500 source faces and 500 target faces to train the model. In general, the source images were rando and pulled out of internet posted videos. For the Zao app, there is a ten second process in which selfies are taken of you in a few different poses: mouth closed, mouth open, raised head, head to the left and blinking. By ensuring that the source images are the “correct” source images rather than random ones, they are able to make that side of the equation much more efficient.
While there is a nice selection of “target” videos and gifs for face swapping, its is still a limited number (I’d guess about 200). Additionally, there is no way to upload your own videos (as far as I could tell with the app running on one phone and Bing translator running on a second phone in the other – the app is almost entirely in simplified Chinese). The limited number of short target videos may simply be part of a curation process to make sure that the face angles are optimized for this process, mostly facing forward and with good lighting. I suspect, though, that the quantity is limited because the makers of the Zao app have also spent a good amount of time feature mapping the faces in order to facilitate the process. It’s a clever sleight of hand, combined with amazing technology, used to create a social app people are afraid of.
The deeper story is that deepfakes are here to stay and they have gotten really, really good over the past year. And deepfakes are like a box of chocolates. You can try to hide them because they are potentially bad for you. Or you can try to understand it better in order to 1) educate others about the capabilities of deepfakes and 2) find ways to spot them either through heuristics or CV algorithms.
Consider what happened with Photoshopping. We all know how powerful this technology is and how easy it is, these days, to fake an image. But we don’t worry about it today because we all know it can be done. It is not a mysterious process anymore.
Making people more aware of this tech, even popularizing it as a way of normalizing and then trivializing it, may be the best way to head off a deepfake October surprise in the 2020 U.S. elections. Because make no mistake: we will all be seeing a lot of deepfakes in October, 2020.
I recently did a talk on deepfake machine learning which included a long intro about the dangers of deep fakes. If you don’t know what deepfakes are, just think of using photoshop to swap people’s faces, except applied to movies instead of photos, and using AI instead of a mouse and keyboard. The presentation ended with a short video clip of Rutgar Hauer’s “tears in rain” speech from Blade Runner, but replacing Hauer’s face with Famke Janssen’s.
But back to the intro – besides being used to make frivolous videos that insert Nicholas Cage into movies he was never in (you can search for it on Youtube), it is also used to create fake celebrity pornography and worse of all to create what is known as “revenge porn” or just malicious digital face swaps to humiliate women.
Noelle Martin has, in her words, become the face of the movement against image based abuse of women. After years of having her identity taken away, digitally altered, and then distributed against her will on pornography websites since she was 17 years old, she decided to regain her own narrative by speaking out publicly about the issue and increasing awareness of it. She was immediately attacked on social media for bringing attention to the issue and yet she persisted and eventually helped to criminalize image based sexual abuse in New South Wales, Australia, with a provision specifically about altered images.
Criminalization of these acts followed at the commonwealth level in Australia. She is now working to increase global awareness of the issue – especially given that the webservers that publish non-consensual altered images can be anywhere in the world. She was also a finalist in the 2019 Young Australian of the Year award for her activism against revenge porn and for raising awareness of the way modern altered image technology is being used to humiliate women.
I did a poor job of telling her story in my presentation this week. Beyond that, because of the nature of the wrong against her, there’s the open question of whether it is appropriate even to try to tell her story – after all, it is her story to tell and not mine.
Fortunately, Noelle has already established her own narrative loudly and forcefully. Please hear her story in her own words at Tedx Perth.
Once you’ve done that, please watch this Wall Street Journal story about deepfake technology in which she is featured.
When you’ve heard her story, please follow her twitter account @NoelleMartin94 and help amplify her voice and raise awareness about the dark side of AI technology. As much as machine learning is in many ways wonderful and has the power to make our lives easier, it also has the ability to feed the worst impulses in us. Because ML shortens the distance between thought and act, as it is intended to do, it also easily erases the consciousness that is meant to mediate our actions: our very selves.
By speaking out, Ms. Martin took control of her own narrative. Please help her spread both the warning and the cure by amplifying her story to others.
