Deep Fake Thoughts

[Source: unsplash.com]

Since the early days of computers, back when everything was made up of vacum tubes and wires, AI was a constant concept to harness. There were many studies and research done, founding the basic theory of AI algorithims that would some day get exploited. But on those days, the applications were simple and rare, not really something people would have at their reach. No, it was all research and demostration in controlled environments, since the early systems weren't capable of large data processing at a reasonable speed.

Decades passed, the transistor was born, chips got ridiculous tiny, bytes became kilobytes, kylobytes became megabytes, megabytes became gigabytes and stuff got serious. Nowadays, thanks to Moor's law (which isn't exaclty a law by the way, just a very accurate prediction), we can process data and incredible speeds and in volumes not even thinkeable back then. AI now ceased to be a toy, and is used in the real world, around us, even if you never realized that. Still, it's not the kind of AI that can become sentient and eventually wipe us all out, no no no... at least for now. No, it's the type of AI that is really good at classifying, detecting, taking an action based on some inputs, learn, and so on. AI can help solve problems in new ways and if well crafted it can be really good at it.

The key point with AI is that instead of writing a program that can solve a problem, you write a program that can be trained to solve that problem. If you are the programmer crafting a program to solve a particular problem, you have to code in all the possibilities around that particular problem, and sometimes that is impossible. Even if done 100% perfect, conditions may change in the future and your program would no longer solve it. So with AI, and more precisly with Machine Learning, you instead program a set of algorithims that can be trained, and you don't have to know the peculiarities of the problem, because the AI will figure it out for you. It's a very different way of doing things, but AI and Machine Learning are hot trends right now, specially because we have the computer power necessary to play with it more extensively, and it's at the reach of anyone willing to try.

One of the recent applications of AI that is getting quite popular is the Deep Fake VFX. Many people are experimenting with it, and results are scary. In short, Deep Fake is a technique of image processing, powered with AI algorithims that can process a batch of faces, and is capable of reconstructing a face from the trainning set on top of a video footage. The result is a very accurate representation of the face learnt from the training set that acts and moves with the gestures of the person that is applied to. In laymans terms, it creates a mask from a set of images that you can place on top of your face in a video. Let that sink in for a moment.

I've been seeing a few videos of people using Deep Fake, mostly for entertainment purposes, and the results are astonishingly good! Of course, the mask is not the entire act, it's just one part of it. To get it even better, the 'wearer' of the mask has to act accordingly to the mask it's wearing, otherwise it doesn't work that well. Mix that with an good voice immitation and you can personificate others. To human eyes, if well done, it's really hard to distinguish, but it has its limitations of course, for example, if you don't have a beard and prepare a Deep Fake of someone with one, it will struggle and the results will be poor. There has to be some similarities on both the mask and the face wearing it to make the illusion.

So, what does this mean in our complicated context of media? Well, pretty much anyone can get the hangs of this technique and publish fake videos around. News media could alter their pictures or footage to put someone else on them. It could be a formidable tool for fake news, since the trend is still growing sadly and it's really dangerous. Public declarations can be faked, compromising footeage can be faked, people can be faked. On the other hand, it can be a great tool for entertainment, it can give us the joy once again from loved people that passed away, it can help us remember them, and it can be funny too. It's both a potential weapon and an entertainer. I have no doubts that soon we might be target of dubous content, maybe in the political area, where everything that can be a weapon will be used as such.

So, yeah, it's a little bit concerning to have this power. But luckily, Deep Fake is not exaclty perfect. So how to battle AI? Well, with another AI of course. On the good side, with proper processing and analysis, footeage can be verified, and another AI can be trained to recognize the flaws, which can also be particularly good at it too, breaking down the deception. It's not all bad news after all. However, that doesn't mean someone will not try to feed us such content, but we need to be critical, as with any media content we consume.

Many applications of AI are still in early stages, a few are operating in the real world like virtual assistants, image processing, or plain experimentation like Google's Deep Mind with its mind blowing image crafting. But tech is advancing, and fast, 20 years ago this type of thing was unthinkeable for the common person, 20 years ago phones were just bricks of plastic with an antenna and a battery, 20 years ago we had Windows ME. We got this far. In the future more and more AI applications will populate our daily lives, making our routines easier, and our jobs simpler. In coming years AI will be more powerful, and maybe scarier, opening new possibilities for both harm and good. Many people have a dystopian view on AI, and for a good reason, but rest assured that sentient AI is still quite far. I'm not even sure I will be able to see one in my lifespan sadly, but we'll eventually get there somehow, and boy it will stir things up.

So what I think of Deep Fake? Well, it fools me most of the time, and the results are really good with little effort, it will be a topic of discussion if it ever gets used in news media with questionable motives, but I think it would be momentarily only, so I wouldn't worry too much about being lied with Deep Fake, we get lied with or without Deep Fake anyway. But I will be particular wary from now on and be more critical on what I see.

Footnote

This article is more of a personal take on the topic, and wasn't intended to be introductory or deep in the concepts. I just tap the tip of the iceberg here. There are far more resources out there about the technicalities and inner-workings of Deep Fake, you're invited to expand on the topic. So please excuse my lack of sources and formality.