Seeing is believing. Well, at least that was the case before we realized that people could doctor videos to propel hoaxes and rewrite history. While we’ve found ways to debunk most hoax images, there is one technological development that is gaining pace so rapidly we may soon no longer know how to tell what’s real and what’s fake.
Deepfakes change everything we thought possible in terms of doctored videos. Here’s all you need to know about them…
What Are Deepfakes?
The term deepfakes comes from a combination of the words “deep learning” and “fakes”. This is because artificial intelligence software trained in image and video synthesis creates these videos. This AI can superimpose the face of one subject (the source) onto a video of another (the target). More advanced forms of the technology can synthesize a completely new model of a person using the source’s facial gestures and images or video of the subject they wish to impersonate.
The technology can make facial models based upon limited visual data, such as one image. However, the more data the AI has to work off of the more realistic the result is.
This is why politicians and celebrities are such easy targets for deepfakes since there is so much visual data available online that the software can use. Since deepfake software is available on open-source platforms, people on the internet are continually refining and building upon the work of others.
The Origins of Deepfake AI Technology
The technology behind deepfakes was developed for a multitude of purposes. Much like Photoshop, the software has professional, entertainment, and hobbyist uses. And just like Photoshop, despite the creator having no malicious intentions in creating the software, this hasn’t stopped people from using it for malicious purposes.
Face-swapping technology was initially mainly used in the movie industry. One of the most famous instances is in the 2016 film Rogue One: A Star Wars Story. In the movie, filmmakers used face-swapping and video synthesis technology to recreate the character Grand Moff Tarkin. A younger version of Princess Leia was also created in the film. In both instances, models of the original actors’ faces were superimposed onto stand-in actors.
Apps like Snapchat also use face-swapping technology to create fun filters for users. The developers behind these apps continually refine face detection and tracking to apply these filters more effectively.
Meanwhile, others have developed video synthesis tools to create holograms for educational purposes. For example, one project developed video and facial synthesis software so that the testimony of Holocaust survivors could be presented as interactive holograms at a museum.
Why Deepfakes Are Making People Nervous
When people realized that scammers and people playing hoaxes used Photoshop to create fake images, we had to become more skeptical about what we considered proof. Luckily, there were many ways to detect whether an image was fake, even with the naked eye.
In addition to this, creating a convincing doctored image in Photoshop is relatively labor-intensive. Not just anyone can slap together two images and make them look realistic.
But deepfakes are different. Machine learning makes life easier, but in this case, it makes fakery significantly easier. Firstly, the software is widely and freely available. FakeApp, for example, is a popular choice for creating deepfakes. You don’t need advanced skills to apply a face-swap, the software will do it for you.
Since AI and deep learning help create deepfakes, the tech also improves and becomes more convincing at an alarming rate. It won’t be long before these edits aren’t visible to the naked eye.
In a world rife with fake news, convincing deepfakes could prove to be a chaotic force against what we believe to be true.
The rise of deepfakes is also taking place at a time when AI voice synthesis is advancing quickly too. Not only can AI generate fake videos, but it can also generate voice models for people.
This means that you wouldn’t need an impersonator to make it sound like a politician is making an outrageous statement. You can train AI to mimic their voice instead.
The Consequences of Deepfakes
People already use deepfakes for malicious purposes. People often used FakeApp to create fake videos of celebrity actresses engaged in adult content.
Gal Gadot, Daisy Ridley, and Emma Watson are just a few of the actresses targeted by fake adult videos. These deepfakes swap actresses’ faces into videos by adult film stars.
While several platforms and certain adult websites have banned these types of videos, more appear each day. In fact, some websites specifically create deepfake celebrity adult videos based on user requests.
In most countries, no laws deal with this kind of content yet, making it difficult to control.
While we’re still some way away from the dystopia ruled by misinformation and false video evidence that we see in movies like The Running Man, we are already all too familiar with the effects of fake news.
Deepfakes can be a powerful tool in spreading misinformation. No one has been framed for a crime or faked their death using deepfakes, but what happens when it becomes difficult to tell which videos are truly real?
The consequences of deepfakes used for political purposes is two-fold. Firstly, it makes fake news much easier to spread. Videos are more likely than text or images to convince people that something fictitious actually happened.
People already believe headlines from fake websites with no evidence backing their story. Suddenly, fake stories will have “evidence” showing politicians confessing to wrongdoings or making outrageous statements.
On the other hand, deepfakes could also embolden politicians when dodging accountability. They could always easily claim that an audio or video recording is actually a deepfake.
How Are We Fighting Deepfakes?
While many tech companies are taking their time to reign in deepfakes, a variety of people are developing tools to combat malicious fake videos. AI can fight hackers and cybercrime, but it’s also useful for detecting AI tampering in videos.
The AI Foundation created a browser plugin called Reality Defender to help detect deepfake content online. Another plugin, SurfSafe, also performs similar checks. Both these tools aim to help internet users discern fact from fiction.
Fact-checking websites like Snopes also expanded to calling out doctored videos. But they don’t yet have the tools to automatically detect deepfakes.
Even the US Department of Defense invested in software to detect deepfakes. After all, what would happen if a convincing video of a world leader appeared online, declaring war or a missile launch against another country? Governments need tools to quickly verify the legitimacy of a video.
Machine Learning’s Unintended Consequences
There’s no doubt that AI technology and deep machine learning improve our lives in many ways. But the technology also has unintended consequences.
While bad data is a major hindrance to machine learning algorithms, the human element also plays a role. It’s difficult to predict how people may use certain technology for malicious purposes. You can find out more about machine learning and past mistakes in our guide to machine learning algorithms and why they go wrong.