Earlier this month, we all had a bit of fun when a YouTuber made a trailer of ‘The Matrix’ with Will Smith re-imagined a Neo. It looked very real, & while that was just a bit of fun, the technology – known as deepfake – has not always been used in a positive way.
Concerns have been raised whenever the technology has been abused. For example, when several deepfake porn videos emerged Online featuring well-known celebrities like Taylor Swift, Emma Watson & Gal Gadot in explicit situations. Deepfakes have also been used to depict high-profile famous faces such as Donald Trump, Barack Obama or even Mark Zuckerberg very convincingly speaking controversial subjects, even though it isn’t really them.
But before we go deeper, let’s just try & understand what are deepfakes, & how AI is being used to manipulate & superimpose images & videos.
What are deepfakes?
A deepfake can best be defined as a portmanteau of ‘deep learning’ & ‘fake’, a technique that applies neural net simulation to data sets to create a fake. For you & me, it simply means deepfakes are fake videos, images or audio recordings that look & sound just like the real thing.
Deepfake is all about combining & superimposing existing images & videos onto source images or videos using a machine learning technique with the help of artificial intelligence (AI). The latter learns what a source face looks like from all angles, & will then superimpose it on the original or target image or video, just like a well-made (digital) mask.
With the use of Generative Adversarial Networks (GANS) AI algorithms can be used to create such fakes. While deepfakes can be used to create digital media that are harmless pranks, it can also be used to generate fake news or malicious hoaxes.
Deepfakes that shocked the world wide web community
Apart from examples already provided in this piece, a few recent deepfakes took the www by storm & shocked audiences:
- Hollywood & Bollywood celebrities: Top American & Indian film actors have become unwitting victims of the technology as scores of pornography sites have hundreds of videos featuring stars like Scarlett Johansson, Priyanka Chopra & Deepika Padukone.
- Talking Monalisa: While this was not done as a prank or with a malicious intent, researchers of Samsung’s AI Lab & Russia’s Skolkovo Institute of Science & Technology generated a “photorealistic” talking head model using adversarial learning. It was part of a new research to shows it is possible to generate realistic content even with rare source image.
- Bill Hader morphing into Tom Cruise: Another creator of deepfakes, that goes by the screen name of Ctrl Shift Face shared altered videos inserting Jim Carrey into The Shining, & Sylvester Stalone into Terminator 2. He even used a 2008 clip of Saturday Night Live star Bill Hader seamlessly transforming into Tom Cruise who he was impersonating during this interview.
How dangerous are deepfakes?
Swapping out one face for another or “photoshopping” images to suit one reality has been on since a long time. The manipulation of digital videos & images is not new. But the advancement in technology & AI is making such manipulation to be much more believable to the extent that it is becoming harder to identify fake from real.
In 2017, an anonymous Reddit user named deepfakes doctored a porn video with the face of Wonder Woman star Gal Gadot. The user was soon banned by Reddit, but the technology was out & about.
Software tools like FakeApp & DeepFaceLab are available to everyone. While these apps may not do what the technology does, they are pretty good at improving & repairing videos, try on clothes or hairstyles, & so on. It will not be surprising if they end up being used for nefarious purposes.
Deepfakes are not just used for pornographic reasons. The ill-effects of this technology is far reaching & of course, outrageous. It can easily be used to spread false information or fake news or even fabricate media in a fashion to alter people’s memories of particular events.
Highly advanced technology driven by AI & its subset, machine learning makes them easier to create deepfakes, & as difficult to detect.
However, the real danger is that it will make us question everything on the world wide web or even in the real world. Whether what we see in news, for example, is real or “deepfaked” is a question that is not far away. And it is nearly impossible to prevent deepfakes from being created or to prohibit them from spreading on social media & elsewhere.
A recently shared doctored video of politician & US House Speaker Nancy Pelosi is the most recent example on why we should be worried about such an advancement in technology. In the video, which was low-fi even by recent tech standards, she appears to be slurring her speech & repeating her words, making her appear confused & drunk. It was even tweeted by US President Donald Trump and has gathered 2.5 million views on Facebook.
Combating deepfakes
Even as social media is being used a vehicle to spread these deepfakes, it is the same medium that is also working towards detecting & eliminating deepfakes.
In the fight against deepfakes, Facebook, Microsoft, the “Partnership on AI” coalition, & academics from 7 US universities have now launched a contest “to encourage better ways of detecting deepfakes.
The Deepfake Detection Challenge, as reported by this Website earlier, will run from late 2019 until the spring of 2020.
Experts say that there are 2 ways to combat deepfakes. One is the use of technology to detect fakes & legal framework that will help take down fakes & punish those who make deepfakes with malicious intent. However, these are just talking points for now. The “How” of it seems to be far away.
However, till the time there are strict regulations on combating deepfakes, here is what you can do to stay safe:
- Make your accounts private. Limit what you share on the world wide web, especially the stuff you know can be misused later.
- Regular searches of your name or info related to you on the internet can lead you to any such misuse of your personal details. If you find it, you can then work on removing it from the internet.
- Take action & reach out to authorities if you find your details being used illegitimately. Be it the local police or cyber crime divisions in your respective states or countries, calling attention to hoaxes is the first step to avoid unbecoming situations.
