Thanks to Deepfakes, Reality Is Now an Illusion

I first began pointing out the potential of deepfakes a few years ago. Since then, the internet has been flooded with them, and while we have seen predicted their use for political ends, the underlying reality is that as recent research shows, in 96% of cases they are used to degrade women.

What’s more, technology does not stand still, and it continues to produce ever-more remarkable results by lowering the entry barriers to tasks once the preserve of experts: algorithms that will write any kind of text are now within anyone’s reach.

The same is true for images and videos. A well-written article in Ars Technica, “AI image generation tech can now create life-wrecking deepfakes with ease”, illustrates how this kind of manipulation can be done more and more easily, meaning that some people now spend a lot of time deleting images of themselves on the web.

The authors of the article started out using images of volunteers, but changed their minds when they realized the potential harm to their reputation, and instead created a few images of a fictitious person in various situations. From there, they created other, potentially compromising photos and videos. One of the is John, a school teacher. Starting from seven images that could be found on any social network, it asks an algorithm to generate images of John naked in public, next to a naked woman in an office, dressed as a paramilitary guerrilla fighter, wearing an orange prison uniform or with a facial expression that could indicate pain or pleasure, among others. With the right text, any of the images could be strongly incriminating, and lead to all kinds of professional and personal consequences for John.

To do this, the authors used Stable Diffusion version 1.5 and a deep learning generation model called Dreambooth that uses a few images of a person to create images in all kinds of contexts. The images can come from anywhere: from social networks, or even from frames captured in a video.

Algorithms are now easily able to distort our perception of reality. Do you want to make someone younger or older? Disney, after having to do it through expensive manual post-production methods, having previously sought younger or older actors who resemble the protagonist, has now created an algorithm trained with images of actors over their careers.

Technology, as always, is neutral, and can be put to good or bad use. In any event, by now there’s no point in trying to remove our images from the internet: the task brings the Myth of Sisyphus to mind, and would be a waste of time for anybody with even the smallest public profile, and given the times we live in, would merely heighten our sense of paranoia.

Instead, we need to educate younger generations to be more responsible, while questioning “proof” on the basis that it may be the product of manipulation that once required a high degree of skill, but that can now can be done by anyone. Until society as a whole comes to terms with the sad reality of deep fake, if it ever does, expect the worst and just try to keep it real.

 

 

—

This post was previously published on MEDIUM.COM.

 

***

 

—

Photo credit: iStock.com