A combination of the phrases “fake” and “deep learning”, falsified videos a.k.a. deepfakes first made their way on the internet in late 2017. Powered by generative adversarial networks (GANs), an innovative deep learning method for introducing a convincing digital “face-graft”, where the expressions of one individual are morphed onto another.
For most of the general public, the first exposure to deepfakes occurred back in 2017, when a Reddit user posted videos that displayed celebrities, like the incredibly beautiful and talented, Scarlett Johansson, in compromising sexual scenarios. Of course, the videos were not real, but a combination of a porn actor’s body and celebrity’s face fused via deepfake technology.
Initially, mostly celebrities were target to such charades, since deepfake algorithms require a considerable amount of video footage for creating a deepfake. After some time, politicians too became victims to deepfake videos, stirring quite the controversy. Several deepfakes have gone viral over the years, with some believing they were actually real.
Bill Hader morphed into Al Pacino and Arnold Schwarzenegger, while mimicking their accents and way of talking. Mark Zuckerberg openly admitted that Facebook’s true goal is to exploit and manipulate its users. And, President Obama used a swearword to describe the now president, Donald Trump. As you can see, the results of these manipulated videos are amazingly convincing.
- These are the data breaches to watch out for in 2020 (opens in new tab)
A nuisance to society, elections and politicians
According to a report from startup Deeptrace, at the beginning of 2019, there were approximately 7,964 deepfake videos online, which shot up to 14,678, nine months later. As of June 2020, a new study from Sensity reveals that almost 49,081 deepfake videos have been uploaded online, showing an increase of more than 330 percent, since July 2019.
Most of these deepfakes play with the perceptions of society, while others tend to serve “humorous” purposes. For instance, the famous YouTube creator “CTRL Shift Face”, morphed Donald Trump into a character from “Better Call Saul” known as “Jim McGill”. In the scene, the now President is shown describing the details of money laundering.
The video is a straight up comedy, and it did not have any serious implications. However, another deepfake of Trump did spark quite the outrage. In a video posted on Facebook, Donald Trump is found provoking and mocking Belgium for remaining in the Paris Climate agreement. Although the video was fake, many people though it was real, and continued to insult Trump.
One of the biggest deepfake conspiracies in recent memory of course is from a commercial debuted by State Farm, which shows footage from 1998 of an ESPN analyst, who is making extremely accurate predictions and foresights about the year 2020. Almost everyone believed it was real, but the clip was not genuine; surprising and amusing viewers all over the world.
Damages the poise and character of celebrities
One of the worst implications of deepfakes is that it completely damages the dignity and reputation of celebrities. Deepfake pornography is actually a thing, and it prominently began surfacing, back in 2017 on platforms like Reddit. Among the most discussed and sought after Deepfake, included that of Daisy Ridley, which even made its way on several articles.
After this incident, the internet exploded with pornographic deepfakes of various other celebrities. A report published in October 2019 even revealed that 96 percent of all deepfakes were pornographic in nature. One can already imagine the toll it could have taken on celebrities and other popular individuals, who were being targeted for obscenity.
Scarlett Johansson, a recurrent victim of deepfake pornography, even spoke to The Washington Post publicly in December 2018. She highlighted that although deepfakes did not tarnish her public image, they do have greater potential to threaten women of lesser prominence, who could have their reputations damaged, due to depiction in involuntarily revenge porn or deepfake pornography.
There was even a Linux and Windows application called “DeepNude”, which would remove clothing from images of women. A lot of individuals were victim of blackmailing due to this app, which had both a paid and unpaid version. The paid one costed around $50, but the results of it were quite realistic, leading to the apps removal. However, many versions of it still continue to exist.
- Eight AI trends we’re watching in 2020 (opens in new tab)
A reduction of trust in news circulating online
In addition to changing our perceptions, tarnishing the reputation of celebrities, targeting victims for blackmailing, and influencing voting behavior, deepfakes are now eliminating the trust in news circulating online. Since deepfake technology is improving faster than many believed, the “realness” of such fake content is becoming more and more convincing.
Many deepfake videos can not only manipulate facial expression, but they can also now perform a myriad of movements, including head rotation, eye gaze and blinking, and full 3D head positions using generative neural networks. Never one to miss an opportunity, Google has claimed that its working on a system that has the ability of detecting deepfake videos.
For this, they are creating deepfakes themselves, as stated in a blogpost on 24th September. They have created a large dataset of 363 real videos of consenting actors and an additional 3,068 falsified videos that will be utilized by researchers at the Technical University of Munich. Regardless, deepfakes already have made their mark and eliminated the trust factor of news circulating online.
Among the most targeted sectors for deepfakes, divided by percentage include 62.7 percent by entertainment, 21.7 percent by fashion, 4.3 percent by sport, 4.1 percent by business, and by 4.0 percent politics. And, currently the most targeted countries include USA and UK (making up about 61 percent of the majority targeted), South Korea, India, and Japan. Of course, these numbers will keep increasing over time.
Wrapping things up
One thing is clear that fake news will be circulating a whole lot more, across the world. And you can expect a lot of individuals harnessing full use of these technologies. Pretty soon, malicious actors may also adapt their techniques to force people into revealing private information, which is a risk to your online anonymity; one that cannot be resolved through the use of VPNs either.
With deepfake technologies improving on a daily-basis, it is crucial to remember that you need to assess the authenticity of videos and photos going viral on the internet. This means, you need to do your due diligence in identifying fake news, before sharing it. At the same time, you need to assess the social context in which a deepfake was discovered to understand its precise purpose.
- 2020: preparing for the evolving security landscape (opens in new tab)
Muhammad Hamza Shahid, Online Privacy/Security Advocate, BestVPN.co (opens in new tab)