DeepFakes, DeepNudes and... YouTube?
With the prominence of technologies that can create photo-realistic fake images of people like DeepFakes or modify any image of a person as if they were naked like the now infamous DeepNudes. It's fast becoming ever easier to imagine what a future powered by machine learning can look like. This future of self driven cars and even machine generated music or art is the stuff of dreams for some and nightmares for others.
But in truth, in a slightly less glamorous fashion, technologies and techniques like GANs (Generative Adversarial Networks) while invented and named in 2014 have been in existence in a far less obvious manner for over a decade in the media space. Simply (and crudely) described, Generative Adversarial Networks (GANs) describe a class in machine learning where computers learn a certain task by having two algorithms compete against one another both improving as they compete (if this sounds like any sports match you've ever watched it's because it is.)
A more specific example, if we wanted to teach a computer to draw a cat sitting on a porch, we'll have one algorithm that tries to draw a cat sitting on a porch and the other algorithm trying to discern whether the drawing is a cat sitting on a porch. With our example, as the cat drawing algorithm gets better so will our cat recognising algorithm.
Now that we're armed with the new found knowledge of what GANs are, we can go back 14 years to the early days of a small platform called YouTube. While we all know what YouTube is, some might not know what the "YouTube Algorithm" is meant to do. The algorithm in question, is designed to keep users on the site watching videos for as long as possible and the longer they stay, the more revenue YouTube generates.
What does all of this have to do with GANs you are asking? Well, you see, for a GAN to work you need two competing algorithms, in our YouTube example, one is the actual YouTube Algorithm designed to keep you on the site by serving you the best content for you. On the other side of that algorithm fiercely fighting the good fight to make sure their content makes its way up the trending watch list is no other than the content creator.
For almost a decade and a half, YouTube has been training it's algorithm to serve you better content while content creators have been training to beat the algorithm via different means. Sometimes flat out tricking the algorithm using unscrupulous methods, however, ultimately, improving the quality of their content. Otherwise, they wouldn't get those coveted views and exposure for their hard work. That's the very essence of a GAN.
It's all too easy sometimes to fear the unknown and envision the most dystopian of futures when thinking of these complex technologies that can achieve what often is perceived as mind blowing, indistinguishable from reality, fake nude photo of a real person. The truth, if past experience is anything to rely on, is that these technologies will make us better at everything they touch as we learn to work with them to improve all aspects of the world around us.
Just look at the YouTube content creators, is it frustrating for them to deal with the shifting winds of the YouTube algorithm? Sure. Is the content generated today better than the content generated 10 years ago? Undoubtedly.
While being careful not to mix correlation with causation as there are many other factors at play here and those familiar with the YouTube ecosystem know all too well how the changes to the algorithm single handedly killed channels that have very short content, no matter how good it was. However, by and large, people spend more time on YouTube, which suggests we're enjoying the content more today than in the past.
Should we be concerned at all? Yes. With the emergence of any new technology, we'll have to adapt and learn to live with what new reality it creates. But really all we need to do is make sure it's a bright one we want to live in. After all, there is no turning back time or putting the Genie back in the bottle. Personally, I can't be more excited. Just thinking of possibilities and innovations already applied to medicine in the discovery of new drugs, photography, art, music and so many other fields are just a small reminder of what a great time this is to be alive.
Nadav Yogev - IT Director