Deepfake dangers and how people are fighting back

Most of us have probably heard of deepfakes, which have been widely reported about in the last few years. As developments in AI have made it possible to generate artificial content to an almost limitless degree, it’s possible to imitate nearly anyone in photography and video.

Already movie studios are proposing to capture images of actors to use in the future without needing the actors on set (one of the main concerns fueling the SAG/AFTRA strike). But deepfakes can also be used to manipulate people’s opinions or in various scams.

Let’s take a look at the issue of deepfakes and what is being done to fight back.

What are deepfakes?

You may recall photos of Pope Francis looking a bit too cool in a large puffer jacket earlier this year. These images were created using AI software Midjourney, but as they weren’t so unusual to be immediately recognized as fakes, many people shared and commented on the photos, believing that the religious leader had actually taken on a new hip-hop image.

The pope photo is a relatively innocuous example of a deepfake. The term ‘deepfake’ comes from the complex deep-learning architecture in the generative AI programs that create them. These fake videos, photos, or audio files can be used for various purposes, including market manipulation, creating pornographic material featuring individuals without their consent, impersonation for fraud, and political misinformation. This means there are many crimes that can originate from this one technology. 

As the pope’s example demonstrates, deepfake technology has great potential to misrepresent particular individuals for any number of purposes, for entertainment and malicious purposes alike. 

In incidents that show the potential hazards of deepfakes, last year, a video of President Zelensky of Ukraine telling his soldiers to surrender appeared on a Ukrainian website, though this was almost immediately recognized as a fake and debunked. In the US, faked images appeared, imagining a physical struggle between Donald Trump and the police that didn’t happen in real life. Also troubling was a faked video of former Speaker of the House Nancy Pelosi that made her sound possibly ill or drunk. Like the Zelensky video, both of the US examples were quickly debunked, but not before many detractors spread the content to others. 

These are just some of the fakes of well-known people that have popped up in recent years. The issue is that anyone filmed, recorded, or photographed is potentially vulnerable to an identity attack. 

Scams using deepfakes

Real-time analysis of deepfakes may be just what is needed. According to the Los Angeles Times, real-time deepfakes are already being used for the purposes of fraud. In Newfoundland, a man has been accused of using false audio to scam $200,000 from grandparents who believed they were hearing from a family member. 

“Catfish” or romance scams are also on the rise, and the Federal Trade Commission (FTC) reports losses of $1.3 billion due to cybercriminals posing as love matches for their victims. 

This kind of scam more commonly operates through email, online chat, and photos, but it is also possible for the scammers to speak to their victims in real-time using a convincing deepfaked audio rendition of their real voices.   

Cybersecurity experts now suggest using codewords when talking to family members about financial matters, to prevent any costly deceptions. Beyond that, it’s important to stay up-to-date on the dangers of deepfake technology. 

How to detect deepfakes

As with all types of digital content, now we live in a world where it’s increasingly difficult to notice when we’re engaging with AI-generated content. Can you spot the bot? The answer is quite possibly not. So it’s always best to stay aware of the techniques and risks. 

AI-generated images often show tell-tale signs that they’re not real, such as deformed human hands, which have been difficult for AI programs to reproduce. But as the Washington Post reports, generative art is improving, so it may not be long before it’s no longer possible to distinguish between real and artificial imagery and audio. 

Deepfakes can be truly convincing to the untrained eye. So do we have any method for finding the fakes?

According to Intel, there is. The company has developed the first real-time deepfake detection platform, FakeCatcher, which it claims has a 96% accuracy rate. 

Rather than identifying what’s fake in a video, the platform looks for real elements, such as a heart rate. This can be shown by photoplethysmography, the process through which human veins change color based on oxygen content. This is not visible to the eye, but computer systems can detect it. Then deep learning is used to determine whether the video is genuine or not.

As AI use continues to spread and more people learn the technology, the prevalence of deepfakes will undoubtedly grow. Hopefully, tech companies can keep ahead of the trend and build better means to detect deepfakes before they get out of control.

To learn more about the ways AI and other tech developments are changing the world, check out our weekly Tech Roundup.