Deepfakes are everywhere, so we’ve written a guide to help you learn more.
Can you believe everything you see? The trend of machine-learning powered deepfakes requires us to ask this question every day. But what is a deepfake and why do they matter? In our guide to deepfakes, we’ll explore these questions, and more including where deepfakes come from, how to spot them, and what the future may hold.
What is a deepfake?
A deepfake is an artificial image or video created by a special kind of machine learning called “deep” learning (hence the name). Good deepfakes are created using two algorithms (A & B). Algorithm A is fed real images and trained to create the best fake image possible. Algorithm B is trained to detect whether an image is fake or not. Then, the algorithms feed each other, each getting better at their respective tasks.
The result? Algorithm A becomes so good at producing fake images, it can trick humans.
Going deep
Deep learning is a kind of machine learning. You feed an algorithm examples of a thing and the algorithm learns to produce something that resembles the examples.
Deep learning tries to replicate the way humans learn, using complex node layers to create better, more effective outputs faster.
How is a deepfake different from photoshopping?
Airbrushed models and photoshopped bodies that are thinner or more attractive have been common in ads for years. The latest smartphones use AI so you can turn 5 photos into one to create the ‘best take’.
But, how are deepfakes different?
While the technology that creates the above examples is the same, the images that are produced don’t try to reflect reality – it’s either art or a representation of an idea. What makes deepfakes different – and dangerous – is their intent. Deepfakes are used to present a version of reality that fits with the creator’s intent.
Who makes deepfakes?
Lots of people make deepfakes. From academics and researchers to amateur enthusiasts and institutions. Even film studios use deepfakes to create scenes with actors who have died, or improve the quality of dubbing in foreign-language films.
Why do deepfakes matter?
Today, the vast majority of people get their information about the world and formulate opinions based on what they read and watch online.
At a personal level, deepfakes can lure people into scams and fraud, or ruin reputations by making it appear that someone is doing something that they wouldn’t do. At a global level, deepfakes of world leaders stating made-up claims could incite political destabilisation or even conflict.
“It wasn’t me!”
The very existance of deep fakes creates a one-size-fits-all plausible deniability. They provide cover for anyone to do anything they want because they could dismiss any evidence of wrongdoing as a deep fake.
What are deepfakes used for?
Deepfakes of the average person are likely to be related to bullying or fraud. By creating evidence of false events, it’s easy to place people into ridiculous, dangerous, or compromising scenarios – particularly for women.
Non-consensual pornography accounts for 96% of deepfakes currently on the Internet. Most target celebrities, but there is a growing number of reports of deepfakes being used to create fake revenge porn.
By creating a deepfake of someone’s voice (such as a loved one), it’s easy for fraudsters to gain access to bank accounts or simply trick people out of money with a phone call.
Corporations also worry about deepfake fraud. Identity fraud using deepfakes can lead to share price fluctuations, fraudulent online payments, or security breaches.
And of course, governments fear deepfakes because they help spread disinformation. World leaders, including Sandu, Zelenskky, Putin, and Biden, have been targeted by this technology.
But deepfakes aren’t always malicious. Many can be entertaining, helpful, and even a comfort to those in need. For example, voice-cloning deepfakes can restore people’s voices when they lose them to illness. In America, a Dalí Museum has a deepfake of the painter who introduces his art and even takes selfies with visitors.
Examples of deepfakes
Deepfakes have been used to create evidence of things that never happened. Celebrities have found themselves the unwitting stars of pornography, and politicians have appeared in videos on Telegram saying things that they never really said.
Here are just a few examples:
How to spot a deepfake
Spotting a deepfake is getting harder and harder, but there are ways to identify them if they’re cheaply created.
Trust your gut. If it looks or feels uncanny, verify the image/video.
Bad lip-syncing and patchy skin tones are easy ways to spot cheap deepfakes. There also might be flickering around the edges of transposed faces. Fine details, such as hair, are particularly hard for deepfakes to render, so fringes are another giveaway. Teeth, belts, and shoelaces also have details that cheap deepfakes struggle to recreate convincingly.
But the better we get at spotting deepfakes, the better they’ll become (Algorithm B feeds Algorithm A, remember?). It’s an arms race that will continue to ramp up, but one that we’ve seen before – in the world of computer viruses. Therefore, it wouldn’t be surprising to see “deepfake detector” software, similar to antivirus programs, on our computers in the future.
Will we own our faces in the future?
So, what does the future hold for deepfakes? As discussed, they’re not going to go away any time soon. And it’s likely that they’re going to become more real and harder to detect. More tools will come along to help us identify what’s fake and what’s not.
The content that feeds deepfake algorithms is being pulled from all over the internet. As we saw in the examples above, if you have a significant online presence, you could be a target of deepfakes.
For now, the best thing to do is to remain sceptical of videos you see on social media that appear to show something that seems unlikely. Trust your instincts – if it looks uncanny, or sounds wrong, it probably is.
Sources
⦁ https://security.virginia.edu
⦁ https://spectrum.ieee.org
⦁https://en.wikipedia.org/wiki/Deepfake
⦁https://www.theguardian.com/technology/2020/jan/13/what-are-deepfakes-and-how-can-you-spot-them
⦁https://us.norton.com/blog/emerging-threats/what-are-deepfakes
⦁https://withpersona.com/blog/what-are-deepfakes
⦁https://www.columbian.com/news/2024/mar/24/one-tech-tip-how-to-spot-ai-generated-deepfake-images
You must be logged in to post a comment.