Posted by Matt Ellwood
In December last year, a Reddit user called DeepFakes created a whole new Artificial debate by creating and releasing a self-titled code package called DeepFakes. The software allowed the faces of famous actresses to be swapped with the faces of porn stars, allowing Reddit users watch celebs have sex. The result was so realistic, any watcher would think that Gal Gadot, the lead in Wonder Woman, was now performing sex scenes on camera or that Emma Watson had run out of money and was stripping off for cash. The porn industry is a pioneer in new technology, embracing motion controlled porn, sex avatars, biofeedback and 3D porn. The technology has implications on how the advertising industry uses talent, license personal brands, and the trust consumers will have in video content.
DeepFakes are generated by a deep learning algorithm that is fed thousands of images of a person, all from different angles and doing different expressions. The algorithm then learns how to reconstruct the face, place it over another person's face and then save out image frames. The new face is not a copy from one of the learning images, it learns from those images about how the face reacts and looks, and then predicts what it would look like from different angles - like a creepy, modern day remix. The more pictures that the system is trained with, the better the result. Sven Charleer, a postdoctoral researcher at the Computer Science Department, Augment Lab in Belgium thinks of it like this: “imagine if you could stare at someone for 12 hours straight, observing all their expressions and absorbing that into your brain. Then that person asks you to sketch his face on paper, smiling, crying, any expression you’ve observed. It may seem hard for us to do, but for AI, it is a piece of cake.”
Porn has always been a quiet (so to speak) innovation leader. Huggies diapers came out with a pregnancy belt that allowed an expecting father to experience the feeling of a baby kicking inside their partner's womb. This used Haptic technology that was made possible because of the porn industry pushing the technology in its early stages. Computerised orgasms may sound like science fiction, but the adult industry is driving this technology, and it won't be long till it’s mainstream. VR is also another technology that has been propped up by porn. According to a recent Pornhub insights report, since launching its VR category in the Spring of 2016, the number of videos went from 30 with a handful of views a day, to over 2,600 videos and more than 500,000 views a day as of this month. The uptick hit around Christmas last year when VR headsets became the high tech gift of the year. Porn drove the creation of DeepFakes. Celebrity fakes was a large subreddit on Reddit, and a user called DeepFake found a faster and better way of doing it for video. DeepFake then released the code that he’d written, so just about anyone could create ‘fake’ celebrity porn videos.
The implications of AI are far-reaching, but that’s not new information for anyone. Chatbots will take our customer service jobs, marketing will be automated, and we will no longer need the Uber driver to drive the Toyota Prius'. On the flipside, it opens up new possibilities on how we (the advertising industry) use talent. The current system requires the time-limited talent to be on set for the entire shoot. There is also the issue of the necessary talent being in a different country, or having unrealistic accommodation requirements. What if, in the future, talent licensed their image and provided visual training data to production companies to feed into their AI DeepFake system? In this world, a shoot could use a body double who might have an entire week free for shooting, instead of a four hour window with the real talent. We already do this in a more ‘analogue’ fashion with stunt doubles, but with the obvious dilemma of how similar they look front on to the camera. AI ‘DeepFake’ solves this.
With this technology so accessible (all open sourced and downloadable in Python coding language), it’s ready to be abused. People need to be able to trust what they see. But with DeepFakes, that could be almost impossible. Any person with a computer could use Google Images to train their DeepFakes branch with a plethora of Donald Trump, and use it to declare war on North Korea… if they really wanted. Fake news posted on dodgy websites are naively trusted around the world, and that’s just text. Take that, and think how convincing ‘fake’ video is going to be. What will our verifier of truth be? Maybe we’ll end up with a general distrust of video in the future.
Even though porn is probably the most quietly discussed industry on the planet, it is more important than ever to be keeping up with what’s going on there. Reported to be worth a whopping $97 billion plus in 2017, it’s the area with money to invest in pioneering new technology. Face swapping is the freshest development in AI, but it’s just a matter of time until it starts popping up outside of the shady, incognito tabs of the web. So here we have a double-edged sword - while ‘DeepFake’ could transform the way we shoot content, it might also take fake news and distrust in the industry to a whole new level.