Ctrl:The Deepfake Problem

Caio Brighenti, Maroon-News Staff

“I’ll believe it when I see it.” These classic words — universally evoked by skeptics of all varieties — are on a fast track to lose their meaning altogether.

It’s easy to see why this concept has become the barometer of truthfulness. Everything we know and believe is based on our past experiences. We have an intuitive sense of the world and it takes new personal experiences to invalidate the assumptions we carry.

For instance, if someone were to tell me the Colgate cruiser schedule had been massively improved and was now perfectly usable, I’d immediately feel skeptical and reserve judgement until I saw it myself. It’d take more than just second-hand information for me to discard my beliefs on the cruiser; no, I’d need to personally experience it myself before I even let myself become hopeful.

There’s a certain sense of comfort that comes with evaluating new information yourself. People and news sources can lie to you, but surely your own eyes can’t. Unfortunately, today’s explosion of misinformation is ruining even that.

Doctored images have been a thing for a long time. Even before Photoshop wizards were fooling Facebook grandmas, images were being manipulated old-school style by directly altering negatives. Over the years, our trust in photographs, especially those seen on the Internet, has diminished almost entirely. Now we can hardly see an image that’s even just a little strange without considering that it might be manipulated.

But while trust in images has deteriorated, the opposite has happened with videos. Unable to rely on online images and news for fear of doctored images and misinformation, we’ve instead begun to rely on video for our dose of pure truth. Is there anything that feels more authentic than shaky amateur smartphone footage of an event? Single frame images are easily manipulatable, but entire videos complete with audio surely must be trustworthy?

Well, even though the answer to that question has long been yes, Deepfakes are now eroding our last bastion of authentic online media. If you haven’t heard of Deepfakes, they’re highly complex models that rely on deep neural networks to digitally superimpose human faces onto videos of other humans. Basically, Deepfake technology allows you to create highly believable videos of just about anyone doing just about anything.

Initially, Deepfakes gained public notoriety when a publicly posted Deepfake tool began being used to create fake pornography of celebrities. This in itself already demonstrates the danger of this technology, raising not only legal and moral questions about the disturbing idea of artificial adult videos, but also practical issues with blackmail and extortion. But Deepfake technology has far more sinister applications.

Deepfakes, combined with clever voice acting and voice manipulation tools, can create videos of political leaders speaking that are often completely indistinguishable from reality. The more widespread this technology gets, the easier it will be for purveyors of misinformation to sell their hoaxes. You might be skeptical to read that Barack Obama finally admitted to being a communist, but watching and hearing the man himself say it just might be enough to tip you over the edge. Clearly, the potential for wide-spread misinformation campaigns with it is enormous.

Facebook and other major tech companies are funneling major resources into tech-based solutions for this problem. In other words, models to detect Deepfakes. However, these solutions come with an implicit problem: building algorithms to better detect Deepfakes can inform Deepfake creators on how to make their creations more believable. This cycle will continue until every Deepfake is truly indistinguishable.

We’re out of options here. The only one left, regrettably, is to elect to be cynical and finally retire the golden standard for truth. No, we can’t even trust our own eyes anymore.