Disclaimer: This is a personal piece with no intention to hurt anyone. Its a small experiment that was conducted within a few hours. All information has been picked up from public domain. All distortions are for testing and experimental purposes only. The intention behind this post is to demonstrate how easy it is to create a deep fake video.
One of the biggest problems that stands in front of the world with the advances of Artificial Intelligence is that of deep fakes. That is the proliferation of synthetic media.
Events and things that never happened are being created in an artificial manner.
Very simply put, deep fake is output of an algorithm that can manipulate images and videos to embed speech with great precision into a video or image format. When skilfully done, it is difficult to identify the difference in the real and the fake video. The change is so deeply done, that it embeds at the root level. Of course, the subject or the person is largely unaware of the reality or the fact that anything of this sort is available somewhere. They are completely unaware of their deep fake content.
This is very alarming and dangerous. So far, morphing and content distortion was isolated to the pornographic industry where, very disturbingly, videos were being modified and being used against women.
However, manufacturing these fake videos required a certain level of manual precision and labour which was a huge barrier.
Unfortunately, with the advent of AI, this barrier is breaking.
If you haven’t heard about Bakamitai or Dame Da ne, its all right. You can simply google and find out the rage that memes and synthetic content based on YouTube and internet in general.
The proliferation of this content made me wonder how easy or difficult is it to build such content myself. With a little research, I came across this video where the host gives very precise instructions on making fake videos, with the intention of exposing it.
Tempted with the easy with which it is done, I tried out two videos myself. I was able to make them in no time.
The first attempt was with #RaGa. Here is the output of this.
Just to keep things in right perspective, I then prepared another one with political opposition to the gentleman in the first video.
Total effort spent in this activity was less that 3 hours.
The ease with which I could create these memes underscores the greater risk of deep fake exposure that we face in our everyday lives. This was just three hours of my effort. With more effort and some coding, one can make close to real memes and videos and publish them.
With political and other motivations, such videos and capabilities pose a greater risk of knowing the truth. This is great threat and human intelligence is not enough to make out the difference in the two.
I shall spend some more time in my subsequent blog to understand the technical aspects of this process. Perhaps a deeper dive into the inner workings of the algorithms.
Irrespective, this stays something that is really bothersome.