Do you remember that clip where Mark Zuckerberg said that he owned billions of people’s data? Or when Jon Snow apologized for the ending of the Game of Thrones series? There is one surprise for those who believed these things were real: they aren’t. The fake events are generated by using artificial intelligence, known as deep learning, hence the name Deepfakes. You’ve come to the right place if you want to find out more.
What is a Deep Fake?
Deepfake technology is a state-of-the-art tool that can easily stick anyone, alive or dead, existent or non-existent, into a video or photograph that was originally not present in the frame. Of course, the technology has been around for decades now, for example of the resurrection of the late Paul Walker for Fast and Furious 7. But until a few years ago they were time-consuming and took experts and intellect to create. Now, they are simple and easy (although hardly straightforward). This is because of the new automatic computer graphic systems.
The term is derived from a technological term known as “deep learning.” Deep learning algorithms can use data to teach themselves how to solve problems. This is how they create realistic-looking fake media.
How are Deepfakes Created?
Deepfakes weren’t hard to find a few years ago, but machine learning has now made them accessible to everyone. To make a Deepfake, one needs to find the neural network on real footage and provide realistic data and understand what the subject looks like from various profiles and angles and under different lighting conditions. This usually takes a few hours. Computer graphic techniques are then used to create a realistic copy of the subject’s face on the face of another.
Yes, the process has speeded up significantly since artificial technology came into the picture. However, it is still a process that takes time if you want a truly believable copy of the subject. Apart from all that, a few manual tweaks are also necessary to ensure any glitches stay hidden or disappear.
Despite the common assumption that generative adversarial networks control Deepfakes, creating a GAN requires hours of data collection. Additionally, GAN models are not very good at audio and video Deepfakes. They have difficulty keeping up with temporal consistency and keeping the same image aligned in two varying frames.
What are Deepfakes Used for and Who Uses Them?
Non-consensual pornography is where 96% of the Deepfakes are. While most Deepfakes include the faces of celebrities, there has been a rise among revenge porn Deepfakes too.
Yes, women are being targeted, but it would be more apt to say Deepfakes pose a clear threat to people around the world due to bullying, regardless of where they are from.
Additionally, there has been an increase in concerns about audio Deepfakes being used to extort consumers, as identity fraud has become easier by the day. One of the most major concerns was that fraudulent online payments and hacking would become widespread.
To governments, the greatest threat would be overthrown democracies. If it was easy to plant a celebrity in a pornographic video, a politician running for election could just as easily be defamed. An example of this would be Joao Doria, the Governor of Sao Paulo, who is married, participating in a pornographic video.
Because these are unconfirmed points and are rather vague, the ambiguity is, in fact, the greatest danger of Deepfakes. That’s right, the greatest danger is not the capability but what it leaves behind. The existence of Deepfakes provides a shield to anybody for their wrongdoings because they dismiss it as an event that has not occurred. It is a plausible, all-explaining excuse that cannot even be questioned.
Now comes the next question: who exactly is making Deepfakes? Anyone could be the answer: academic and industrial researchers, pornographers, enthusiast enthusiasts, visual effect studios, etc. They are also used by scammers or simply those looking to create havoc.
What Resources are Needed to create Deepfake?
A standard computer is not capable of making a Deepfake of any quality. It is usually done on powerful desktops with highly sophisticated graphics cards or on the cloud.
Not only does it need high-end technology, but it also needs expertise and intellect because you will need to reduce flicker and glitches.
There are a lot of tools available online that help in making Deepfakes, and there are many companies that make these Deepfakes for you. Many apps let you add your face to many celebrity characters pre-defined and trained by the system.
How to Spot a Deepfake?
It comes as no surprise that the better the technology used gets, the more realistic the Deepfakes appear. The problem is that as soon as a new weakness is revealed, it is immediately fixed. For example, initially, the Deepfakes were never seen to blink. As soon as this was pointed out, the creators started developing Deepfakes that blinked.
It is easier to pick out low-quality Deepfakes. Usually, the lip-synching and the skin tone might be off. The edges might be flickering, and several glitches appear. Inconsistent lightening on the reflection of the iris, badly rendered jewelry, hair, and teeth are also signs of a Deepfake.
Facebook has now banned Deepfakes. Governments and universities all around the world are funding Deepfake detection research.
Problems Caused by Deepfakes
Of course, Deepfakes sparks harassment, intimidation, undermining, and destabilization. Despite this, world leaders and army footage Deepfakes will not cause many wrecks since they have their own imaging and security systems.
Yet, many questionable tapes went viral online, such as Elon Musk smoking a joint, which caused Tesla stocks to crash.
Now that Deepfakes cannot even be distinguished from the truth, doubts are sure to raise. For example, Cameroon’s Communication Minister said that the video of the country’s soldiers executing civilians was fake. Some believe him, and some don’t, but that is all part of the Deepfake game now.
Deepfakes are not illegal but depending on the context, circumstance, and content, they can be charged for copyright, defamation, non-consensual porn, and breach of data protection.
As technology becomes more accessible, judiciaries fall into trouble, especially in events where videos can be considered evidence, such as child custody and employment cases. There is also a huge security risk, with Deepfakes copying biometrics and marks that can be used for false identification. The risk of scams is thus increased. For example, if someone texts you for a money transfer, you won’t do it. If a video call comes from relative requesting money, what will you do?
Is Deepfakes Always a Cause for Chaos?
No, they are not. A lot of them are very fun and entertaining.
- There are also instances where Deepfakes were used to voice those who had lost theirs to disease.
- They are also used to represent art.
- Technology also helps the dubbing of foreign films with significantly reduced effort. Also, as mentioned before, resurrection is an occurrence made possible.
Are Shallowfakes and Deepfakes the Same?
Not at all. Shallowfakes are videos that are presented out of context or edited using simple editing tools. While they are not precise and as powerful as Deepfakes, they are still rather impactful. An example would be the incident where Jim Acosta was temporarily banned from the White House. Later, a shallow fake was released that showed the intern jerking the mic off him. This was later discovered to be a sped-up clip of the incident, making it look aggressive.
How to Stop the Offensive Deepfakes?
In 2019, many US states introduced laws criminalizing Deepfake videos of non-consensual pornography and the use of Deepfakes relating to elections. Texas, Virginia, and California have criminalized Deepfake porn, and the National Defense Authorization Act signed the first federal law against it.
Outside the US, the only countries taking action against Deepfake crimes are South Korea and China. In the UK, the law commission is currently reviewing laws to address Deepfakes. The European Union does not see Deepfakes as a major issue.
In spite of the United States’ efforts to legislate in certain areas, these laws do not have any validity. Hence, nobody is sure whether they are implemented correctly and fully.
Many labs have provided ways to identify manipulations using blockchains, watermarks, et cetera. Still, the fact remains that the more manipulative glitches are found, the more measures the video creators take to make the Deepfake look and feel convincing.
Facebook recruited researchers to create Deepfake detectors to make sure its Deepfake ban has properly been implemented. Twitter has taken this one step further, reportedly tagging any Deepfake that is not immediately removed. YouTube said that they would not allow any Deepfakes regarding the elections.
The best way to eradicate Deepfake is using artificial intelligence. Ironical, we know, but effective nonetheless. AI is already helpful in spotting fakes. When it comes to Deepfakes, tech companies consistently try to detect, flag and remove them.