Is It Easy to Create Deepfakes?

Recently, two stories made headlines and raised fundamental questions about the dark side of AI: the fake explicit images of Taylor Swift and a $26 million scam in Hong Kong.

In both cases, the perpetrators used what's known as "deepfakes," meaning a fake image of a person generated by Artificial Intelligence.

I had heard about people impersonating someone else by borrowing their CV and image to pass video call interviews.

Even former French President Francois Hollande was fooled by two Russian comedians who used this technology:

https://www.youtube.com/watch?v=9Lp2wMElw-Q

So, is creating deepfakes within everyone's reach? Is it a criminal offense?

Is Creating Deepfakes Easy?

Honestly, it's terrifying but yes! It's shockingly easy to create deepfakes...

A 2019 study reveals that 96% of deepfakes are sexually explicit images, and of course, the vast majority of these images target women (often celebrities or ex-girlfriends).

I remember my sister telling me a few years ago about "internet tribunals" where middle and high school students would single out a person and put them on trial on the Periscope app. The same thing today can lead to deviant behavior where, "just for laughs," things take a dark turn: targeting a young girl and creating fake images for malicious purposes. Once a photo is put online, it goes viral and it's nearly impossible to completely clean up those harmful digital traces.

It's clear that we have an obligation to prepare properly and understand the ins and outs of the AI revolution in order to face it, because what this revolution has to offer is tremendously positive, but it also comes with its share of darkness.

In practical terms, there are three methods for generating a deepfake:

• "Face Swap" (you swap one person's face with another),

• "Lip Sync" (you artificially move the lips of a face),

• and "Puppet" (you move your own face and facial expressions to animate an artificial image).

The most accessible method is by far "Face Swap." For generating inappropriate images, fortunately you won't succeed using Google. There are very easily accessible tools for blending one person's face onto another image.

But that's not all!

You can create images entirely from scratch using AI these days. While Dall-E and Midjourney have restrictions, Stable Diffusion, being open source, is uncensored. So by using the Fooocus program, it's possible to generate rather inappropriate images of Taylor Swift in 10 seconds.

By the way, Theo had briefly discussed this in his Let's Talk, which I encourage you to watch on the topic of Stable Diffusion:

https://www.youtube.com/watch?v=GiLMJiz63Y8

So yes, it's accessible!

There are even websites that have been trained to generate explicit content using Not Safe For Work (NSFW) prompt instructions, such as "PromptHero."

The Taylor Swift story simply highlights today what has existed for quite some time.

As for the Hong Kong story, that's a different matter entirely. It was a very sophisticated scam requiring much greater resources: a heist for the ages.

What About the Legal Side?

For some time now, there's been talk about the European AI Act, and I invite you to read this article to learn more.

The regulatory framework adopts a risk-based approach, banning AI systems considered a clear threat to people's safety and rights, and subjecting high-risk AI systems to strict obligations before they can be placed on the market. Thus, every creator will need to explicitly state that they are using AI and will need to follow strict rules based on the AI's risk level.

However, enforcing this law against creators outside the European Union is going to be very difficult. It's essentially the same question as with GDPR, where you still have to track down the perpetrators, who are complete unknowns.

The Digital Services Act (DSA) allows individuals to report potentially illegal content, including deepfakes, to online platforms, which are then responsible for handling those reports.

So the legal frameworks exist, but enforcement is not yet effective.

It's therefore up to each and every one of us to be vigilant and careful, especially with more vulnerable populations like our children.

That's why I insist so strongly on the need to understand the world of AI.

Positive Uses of Deepfakes

I deliberately emphasized the negative aspects of deepfakes, but there are positive applications as well.

Deepfakes are widely used in the film industry to, for example, digitally de-age an actor. A few weeks ago, I saw a hologram of France Gall in the musical Starmania.

A few weeks ago, HeyGen created a massive buzz: all you have to do is film yourself speaking in French for 30 seconds, and HeyGen produces a video where you're speaking in Portuguese with the correct lip movements while using exactly the same text.

Technologies similar to deepfakes can be used to recreate the voice of people who have lost their ability to speak due to illness, allowing them to communicate with a voice that sounds like their own.

One Bottom Line

Let's educate ourselves!