What is a deepfake video and how do you protect yourself from one?

Ever come across a questionable video clip of your favourite actor hurling abuses and wondered if it’s really them? There’s no saying they can’t, but if you’ve come across AI-generated content before, you probably have a hunch that something about it is off.

Deepfake videos, as they’re known, use convincing imagery and audio to create simulated content about a person. Apart from being misleading, they are immensely dangerous for the identity that they are impersonating. If current incidents have led you to understand more about this technology, we’re here to educate you.

How do deepfakes work?

Deepfakes use a thousand face shots of two people before swapping them and merging them into one video. This technology is run by an AI algorithm known as the encoder. It learns any similarities between these two faces, and it further trains the video to morph one face over the other. This process also adapts to their facial expressions and, in return, reconstructs the video with a new person.

ALSO READ: To Grok or to not Grok: Everything we know about Elon Musk’s xAI chatbot

Another machine learning technology that helps create deepfake videos involves generative adversarial networks (or GANs). It puts together two algorithms, known as a generator and discriminator, that keep processing cycles of synthetic faces and noise to create realistic faces or expressions.

How are deepfakes made?

While creating a deepfake video today has become simple, most are still created on a high-end laptop. With high-end computing and powerful graphics, these videos can take anywhere between minutes, hours, to days to process.

Some may even require additional touch-ups or effects to reduce glitches or flickers on the output. That said, plenty of filters and apps are available to make convincing deepfake videos today and are effortlessly accessible to anyone with the right tools.

Why are deepfake videos dangerous?

Deepfake videos, although meant for fun, are most of the time intentionally created to harass or demean an individual. In fact, according to a report published by Home Security Heroes, deepfake pornography makes up 98 per cent of all deepfake videos online.

The study even goes on to mention how 99 per cent of people targeted by this technology are women. Apart from swapping face, videos today can also imitate voices, making it challenging to spot one today. 

How does one spot a deepfake video?

Deepfakes can fool just about anyone, making it difficult for you to tell them apart from regular videos. However, the best way to identify an AI-generated video is to look at teeth, nails or lip sync precision.

Despite the availability of technology, it is almost impossible to replicate strands of hair or make the transition appear smooth in a deepfake video. Under any other circumstances, it is best to use an AI-based detector to recognise a normal video apart from a deepfake.

The need to counter deepfakes with a stringent law is perhaps the only way to ensure that they stay out of circulation. While harnessing technology is certainly a move to go ahead with in the future, there is no denying it needs to be worked with responsibly by keeping ethics in mind.