Deep Fakes: Getting Too Real?

What is a Deep Fake?

Ubiquity often results in a noun becoming a verb. We "xerox" a document, or we "photoshop" a picture. Speaking of the latter, you might think the art of retouching or manipulating photos had something to do with the advent of the Adobe Photoshop software in the late 1980s. The truth is, it's been going on since the mid 1800s. The History of Retouching is a fascinating look at the techniques used over the past 150 years, along with some famous examples. Photographic post-processing has long been used in fashion, beauty, art, and political contexts.

The development of software that makes digital image manipulation a point-and-click affair is what moved the process out of the darkroom and into the hands of the masses. And you don't need to spend hundreds on Adobe Photoshop. See my articles Photo Editing Apps and Online Photo Editors to find links to free image editing tools.

It seems harmless to crop a picture, retouch a photo to eliminate facial blemishes, or "accentuate the positive" in a variety of ways. You might start to feel differently about online tools such as remove.bg that make it super easy to remove the background of a photo, or Deep Angel which lets you selectively delete objects from a scene. But when this technology moves into the realm of video and audio manipulation, it gets a little darker, and scarier.

The technique is known as “deep fake,” and of course it’s enabled by artificial intelligence, neural networking, and machine learning. Here’s a harmless example: actor Nicolas Cage inserted into several well-known movie scenes. YouTube has quite a few examples of deep fakes, some good and others easily dismissed.

Creating a deep fake video takes some free, open-source software such as DeepFaceLab, a powerful graphics processing card, and lots of images of the person(s) you wish to fake. Machine learning, after all, requires tons of data from which to learn. Celebrities and politicians provide ideal fodder for deep fakes. But the software described in this video claims a breakthrough, such that only one photo of the target is needed to generate videos that mimic mouth, eye and other facial movements realistically.

Deep Fakes for Fun, Profit, and Deception

A light-hearted deep fake put comedian Steve Buscemi’s head on the shoulders of “Hunger Games” star Jennifer Lawrence. Obviously, deception was not the author’s intent. But a number of actresses have been victims of deep fakes that superimposed their likenesses on the leading ladies of "adult" films.

Actress Scarlett Johansson is one of the best-known deep fake victims. Dozens of bogus salacious clips of her have been viewed millions of times. (Ironically, Ms. Johansson played the faceless voice of an artificial intelligence in the 2013 sci-fi film, “Her”) She was also victimized in 2011, when she was among several celebrities whose private photos were stolen and posted online; the hacker responsible drew a ten-year prison sentence. But sadly, Johannson said to the Washington Post, “The fact is that trying to protect yourself from the internet and its depravity is basically a lost cause, for the most part.”

The popularity of video conferencing has made it possible to deepfake your own face during online meetings, in real time. Avatarify runs on Skype and Zoom, and face-swaps your own face with a celebrity in live video calls.

Perhaps most alarming is the truth that deep fakes have the potential to alter the outcomes of elections. A Seattle TV editor was fired in January, 2019, after a doctored version of President Trump’s nationwide address went out over the air. Several deep fake videos of former President Obama can be found online. It isn’t hard to see how deep fakery could destroy the reputation of a politician or anyone else.

Should We Be Concerned About Deep Fakes?

For the most part, deep fakes are still pretty obviously fakes, once viewers take a good look at them. But the technology is improving at lightning speed as machine learning and artificial intelligence make rapid advances. Many experts are concerned that we will soon find it difficult to tell real video from fake, even when it depicts people we know.

Deep fake tech potentially takes “fake news” to the next level. It further undermines our ability to tell the difference between truth and lies. In such a world, it becomes more important than ever to get our information from trusted sources and avoid sharing every sketchy meme that sparks our outrage.

An article from TechDirt suggests that we shouldn't "go off the deep end" when considering the future impact of deep fake video. We've long accepted the fact that photos can lie, the author says. So "who is to say," he then wonders, "that societal response to deep fakes will not evolve similarly to the response to digitally edited photographs?"

I have my doubts. How about you? Your thoughts on this topic are welcome. Post your comment or question below...