DeepFakes: Real or Fake
“Technology that can make individuals believe as if someone has said or done something when they have not”
This technology can be used to depreciate someone’s public image. The major targets can be politicians to make their fake videos which can be used against them in elections, celebrities to deteriorate their image or females to make fake pornographic videos.
What are DeepFakes?
DeepFakes is a combination of two words ‘Deep Learning’ and ‘Fakes’. It is a subset of AI, empowered with deep learning to create fake videos by moulding images, voice etc.
In simple terms, these are falsified videos which use face grafts, body transfer etc that can be used to spread misinformation about people as if they have said something when they hadn’t said those in real.
How DeepFakes are created?
Anybody who has a computer and access to the internet can technically produce a “deepfake” video, says John Villasenor, professor of electrical engineering at the University of California, Los Angeles.
To create DeepFake videos, it encompasses deep learning models to train the huge amount of dataset. In the process it uses Generative Adversarial Networks (GANs) which allow two models to understand themselves. One model is being trained on the dataset and generate results while the other model check for faults and feed the result into the model then the first model train on data again to minimize the errors. This process goes on until the faults become zero. The bigger the data set is the better the results are generated and appear more realistic.
In similar order to create such videos, preliminary fake videos are produced then the target videos are get scanned and then put in the model to understand patterns in expression and voice modulation. When the model made itself able to understand the patterns, then the new transcript is added to the video and finally, the 3D model of the video is added.
How to detect DeepFakes?
“It is always fun and games until someone gets hurt”. As this technology is new and in its infancy when people play around it for fun but gradually when it would get older and more and more out towards the people it would be more powerful and people may start using it for their bad motives. In such scenarios, it becomes important to recognise whether they are synthesised or original.
It becomes necessary to build a deep learning model which can recognise whether the videos are synthesised. To make a DeepFake detection deep learning model the big tech giants such as Facebook, Amazon, Microsoft etc have jointly launched a challenge “DeepFake Detection Challenge”, in which people around the world are invited to innovate new detection tool which can be used to detect DeepFakes.
Conclusion
As a detector tool for the DeepFakes are still under process and no one knows by when we can expect such technology. Even when we would be having a DeepFake Detector would that be enough? We have seen detection technologies have often lagged behind the creativity.
“Will people be able to recognise the Deepfakes?” The future is still to be let out.