Deepfakes threaten our perceptions and wellbeing

AI-generated video fakes may be the latest insidious threat to online safety. Here's how to spot one.

Monday, 14 November, 2022

When historians are writing the story of this decade, 2022 may be the year when social media finally entered the decline stage of its life cycle.

This is the year of Elon Musk’s botched Twitter takeover, Facebook’s first-ever period of declining user activity, and a growing trend of abandoning social media to boost mental health.

While it remains a positive force in some regards, social media is awash with trolling, artificial representations of supposedly perfect lives, confirmation bias and fake news.

Unfortunately, the latter has now been taken to a new level – known as deepfakes.

These insidious creations are causing huge damage not just to the individuals they target, but also to wider principles of trusting what we see online.

But what are deepfakes? How are they created? And why are they so dangerous?

Fake it til you make it

A deepfake is a video clip in which one person’s face has been superimposed over another, often alongside someone’s voice being dubbed into an existing soundtrack.

People have been superimposing each other’s faces onto still images for decades, creating some of the earliest internet memes in the process.

However, in recent years, AI software has become sophisticated enough to extend this process into three dimensions, across moving video footage.

It takes considerable processing power to project a person’s face across a moving canvas, ensuring it moves and behaves in a semi-realistic manner.

Sound is far easier to manipulate, though it still takes time to marry up vowel sounds with lip movements in a passably authentic way.

The technology underpinning deepfakes originated in the movie industry, but lower software costs and dwindling hardware requirements have democratised its availability.

What are deepfakes used for?

Deepfakes have often centred on pornography, where an unwitting victim has their face supplanted onto the body of someone else.

The intention is usually to cause embarrassment and humiliation to a person whose identity has effectively been stolen, though there’s no legal comeback as there would be for ID theft.

Other leading motivations for deepfake video creation include attempts to smear someone’s reputation or social engineering – both examples of fake news.

Political deepfakes underpinned an attempted military coup in Gabon, and were created in India with different languages dubbed on to candidate footage, to reach the largest possible audience.

Deepfakes have been used in blackmail cases, while identity theft is a growing concern in an age when laptops and smartphones often unlock when they ‘see’ our faces.

Fortunately, there are usually tell-tale signs that a video is false, if you know where to look…

How to spot a deepfake

Firstly, ask yourself why you’re being invited to view a particular video.

Is it politically motivated? Is the person purportedly in the video famous? Are their actions plausible?

Secondly, listen to whether the audio soundtrack marries up. If the volume or background noise keeps changing between one person’s voice and another’s, it’s probably a stitch-up.

We naturally focus on the eyes of people in video clips, but eyes are hard to clone. Look for jerky eye movements, unblinking stares, or people seemingly looking in the wrong direction.

Other giveaways include jerky or unnatural body movements, misplaced shadows and odd lighting, blurred edges around hair and teeth, and a lack of emotional expressions.

It’s historically taken hundreds of still photos to accurately transplant facial characteristics from one person to another.

Less source material means less accurate cloning, so you’re unlikely to become a deepfake victim if you only have one or two personal photos online.

Recent AI advances have reduced the number of photos required for a deepfake to a handful, but this technology isn’t yet widely available.

However, the rising scourge of deepfakes provides another example of the dangers inherent in publicising your life across social media accounts.

The best way to avoid becoming a deepfake victim is to not publish selfies, while ensuring profile photos are either colour-tinted, low-resolution, cropped, or taken at an angle.

Neil Cumins author picture


Neil is our resident tech expert. He's written guides on loads of broadband head-scratchers and is determined to solve all your technology problems!