Although altered photos and videos are increasingly commonplace in our social-savvy lifestyle, the momentum and adoption of “deepfake” content is something to keep a close eye on.
Deepfakes (a portmanteau of “deep learning” and “fake”) are synthetic media in which a person in an existing image or video is replaced with someone else’s likeness.
Perhaps you’ve seen the more entertaining versions of this technology – like Steve Buscemi as Jennifer Lawrence or Steve Harvey as Megan Thee Stallion. Or perhaps you’ve seen recent State Farm and Hulu ads using replacing an actor’s head with another actor’s head. Or maybe a friend shared an Instagram video of their face on Al Pacino’s body yelling “Hoo-ah!”
Deepfakes can be super fun. But there’s a more serious and haunting use of this technology coming, and its widespread adoption and misuse could impact society’s common understanding of things like facts, truth, and proof.
Deepfakes were one of my 10 Digital and Marketing Trends to Watch in 2020, and that prediction is playing out. As machine learning, A.I., and autoencoder neural networks get smarter, more affordable, and more utilized, we’ve seen use of this technology move from academics and huge computers needed to process to a deepfake AI that can fabricate a video clip of you from a single photo.
Not all things called deepfakes are truly deepfakes. Although “deepfake” is the umbrella term to cover this trend, many of the doctored videos we see tend to be “shallowfakes,” which are manually manipulated or edited video that don’t require machine learning or A.I. But for the purposes of this overview, we’re going to use the umbrella terminology of deepfake. Let’s try to make a complex thing a little easier to understand, okay?
THE FUN AND DUMB SIDE OF DEEPFAKES
From the introduction of ElfYourself in 2006 to Snapchat’s FaceSwap in 2016, it can be really fun to put different faces on different bodies. It jars the senses and makes our brains happy. But with today’s deepfake technology, things are leveling up.
To celebrate the 30th anniversary of The Fresh Prince of Bel-Air, a voice actor made this deep fake video of the theme song with seamless transitions from Will Smith to Obama to Chadwick Bozeman to Trump to Arnold to Christopher Walken and more.
And if you like Nicholas Cage, here he is in all kinds of famous movies…
Or here is Steve Buscemi as Jennifer Lawrence. It’s hard to look away…
FOOL ME ONCE, SHAME ON DEEPFAKES
But beyond the fun and games, deepfakes are pretty damn serious and are fraught with risk and negative uses.
Deepfakes really started getting popular and shared around in 2017 with.. you guessed it.. the p*rn industry. That’s a gross path to go down, and you can probably imagine how that technology is being used, so let’s look a little more recently.
One of the most foreboding tactics to illustrate deep fakes is to use this technology to mimic global leaders and make them say things they never actually said.
In early 2018, Buzzfeed partnered with Jordan Peele to share this video where it looks like President Obama says “Stay woke, bitches,” among other things…
A 2018 study by CNAS suggested that deepfakes posed one of the largest threats posed by artificial intelligence.
AI systems are capable of generating realistic-sounding synthetic voice recordings of any individual for whom there is a sufficiently large voice training dataset. The same is increasingly true for video.
As of this writing, “deep fake” forged audio and video looks and sounds noticeably wrong even to untrained individuals.
However, at the pace these technologies are making progress, they are likely less than five years away from being able to fool the untrained ear and eye.-CNAS, 2018
Knowing this trend could impact national security, in 2019 the Pentagon started working to get ahead of it, partnering with CNN to create an educational microsite and campaign teaching you about deepfakes and how to spot them.
What happens if we can no longer trust our eyes or our ears?
For more than a century, audio and video have functioned as a bedrock of truth. Not only have sound and images recorded our history, they have also informed and shaped our perception of reality.
Some people already question the facts around events that unquestionably happened, like the Holocaust, the moon landing and 9/11, despite video proof.
If deepfakes make people believe they can’t trust video, the problems of misinformation and conspiracy theories could get worse.
While experts told CNN that deepfake technology is not yet sophisticated enough to fake large-scale historical events or conflicts, they worry that the doubt sown by a single convincing deepfake could alter our trust in audio and video for good.-CNN, 2019
With the election season upon us, it’s critically important for social networks and media organizations to spend more time identifying, flagging, or removing deepfakes. And that goes for consumers, too. For example, although not a deepfake, this doctored video of Nancy Pelosi appearing to be drunk spread across news feeds, and Facebook decided not to take it down.
Anti-corruption not-for-profit RepresentUs released a deepfake campaign to underscore the fragility of US democracy in the lead up to the 2020 election. These are not real. And you can tell. But barely. Now imagine these were being spread on official or official-looking channels…
You can really get a sense how this could be misused and abused, right?
Closer to home, MIT released this deepfake video of President Nixon breaking the news that NASA failed and astronauts died on the moon during the Apollo 11 mission using “real” footage of Nixon, a voice actor reading the contingency speech, and deepfake technologies to make it believable.
Partially because there’s a lot of forgiveness of 1969-era recordings (TV static, mis-alignment of audio with video), the result is extremely believable. This is just the latest indicator this technology is becoming more accessible and thus more dangerous for our digital future.
Artist Stephanie Lepp created a project called Deep Reckonings that is a series of synthetic, deepfake videos that imagine public figures take responsibility for things in a way we’ve never seen before.
The public figures include: Brett Kavanaugh wrestling with the way he responded to the sexual allegations against him; Alex Jones grappling with his spread of deceitful conspiracy theories; and Mark Zuckerberg confronting his techno-utopianism. These figures don’t admit to actions that aren’t publicly known, but instead take responsibility for actions that are. Deep Reckonings exists in dialogue with the broader conversation about the ethical implications of synthetic media, and how artificial intelligence impacts our understanding of truth. The project seeks not to deceive nor demean, but to imagine and inspire. The videos make their fakery explicit not only to prevent misinformation, but also to leverage a superpower of synthetic media — that we can know they’re fake and they still affect us. In this spirit, Deep Reckonings explores the question: how might we use our synthetic selves to elicit our better angels?-Deep Reckonings, 2020
NOTE: The below content includes references to topics such as sexual abuse that may be triggering to some.
DEEPFAKE APPS ARE HERE AND GETTING BETTER
Increasingly, you don’t need a huge computer and hours of processing time to create a believable deepfake.
In September 2019, a Chinese app called Zao launched in the app store and changed the game forever. Zao made it possible for anyone with a smartphone to make completely realistic deepfakes – like putting your face on Leonardo DiCaprio in movie clips. The app UX and instructions were in Mandarin, so it was a little tough to use for English-only users. However, its launch marked the first plug-and-play consumer-facing app of substantive quality.
Since then, free or ad-supported shallowfake and deepfake apps have increased and gotten more accurate. REFACE, Impressions, and iface are popular face swap apps. Xpression can put your expression on any face. Avatarify creates photo-realistic avatars. Talkr can make any picture talk. Puppets.World can take a video of someone talking or singing and apply to a portrait.
This is from REFACE…
And this is from Impressions…
And here’s a mashup….
The social networks are exploring use of this technology, too.
In December of last year, Snapchat added a new feature that used your selfie photo to deepfake your face over the faces of others. Intended to be used similarly to Bitmoji (for quick, personal responses in Snap messages), it’s like Elf Yourself x Zao, except in one of the world’s most popular social networks.
Then less than a week after that announcement came word that ByteDance, the parent company of TikTok, had developed a feature that called Face Swap using deepfake technology that inserts faces into other people’s videos. The company has now said they will not be launching it publicly. However, given mainstream adoption trends and the fun nature of deepfakes and shallowfakes, it’s likely to be introduced at some point this year.
And memers are in on the action. Folks like @MysterGiraffe and @LemonJezus are using this technology to make highly shareable content. And since there’s no regulatory body policing them, they’re doing it themselves…
Windheim is part of a new group of online creators who are toying with deepfakes as the technology grows increasingly accessible and seeps into internet culture. The phenomenon is not surprising; media manipulation tools have often gained traction through play and parody. But it also raises fresh concerns about its potential for abuse…
Ultimately, regulators need to define what is appropriate use and what could lead to harm.
For now, Windheim is relying on her own judgment to make that call. Before posting her video, she read up on the implications of deepfakes and had a conversation with her colleagues. “We’re never intending our products to help users spread misinformation,” she says, “so we just wanted to sanity-check ourselves.”
In the end, they decided on some ground rules: they would focus their tutorials on making specific memes, never on creating deepfakes outside of that context. As long as it’s entertainment and within meme culture, she says, “we’re in the clear zone.”-Memers are making deepfakes, and things are getting weird, MIT Technology Review
There are deepfake ads, too. Because if there’s a new technology, you can bet some smartie is going to make an ad with it. I’m a little jealous, to be honest.
On the ‘fake human’ side, there’s Ikea Japan, who has released a series of ads and even a live installation featuring Imma—the popular Japanese influencer who happens to be a CGI creation herself. This is an actress with a CGI face.
In more “traditional” deepfake creative, State Farm ran a commercial featuring expertly doctored footage of the longtime “SportsCenter” anchor Kenny Mayne.
Hulu had some complex COVID-19 production issues where their talent couldn’t travel, so they used deepfake technology to make the work…
Lillard has been in the NBA’s bubble in Orlando, and in-person shoots didn’t seem like a good idea for the other star athletes, either. That’s why Hulu shot the commercial with body doubles on an otherwise unoccupied set in Los Angeles, and then used deepfake algorithms to superimpose the faces of the stars into the resulting clip.
The deepfake algorithms were trained on footage of the athletes that was shot exclusively over Zoom, explained Hulu marketing VP Ryan Crosby. “Throughout each shoot, we captured several different facial angles including straight-to-camera, 45-degree angle, 90-degree side angle, looking up and looking down,” he said. “Athletes were asked to say the vowels (‘A’ ‘E’ ‘I’ ‘O’ ‘U’) at each angle along with ‘Hulu Has Live Sports Again.’ We also recorded several different facial expressions that pertain to their specific movements and speaking lines within the spot.”—Protocol, Hulu deepfaked its new ad. It won’t be the last.
Deepfakes are rapidly coming into the mainstream with a momentum that won’t allow for much of a trial period or a break for our collective education gap to keep up. And the combination of deepfakes + A.I. will only rapidly accelerate this — both with deepfakes mimicking real people, and also completely new “people.”
Don’t believe A.I. can generate realistic deepfakes? There’s a website called “This Person Does Not Exist” that generates realistic portraits of people who do not exist, and the results are amazing.
The creator told Mashable he initially created the concept to persuade his friends into believing the authenticity of artificial intelligence and to serve as a warning that A.I. can create extremely realistic fakes – whether to help or harm.
Then there’s Synthesia, a service that enables you to make AI videos by simply typing in text, and sends you a video of a realistic human speaking whatever you typed. Now imagine you can do this for a world leader, or your boyfriend? That technology is coming.
Now imagine you can combine these two technologies. It’s gonna happen…
At CES 2020 this year I made sure to stop and see NEON’s artificial human avatar project — “computationally created virtual beings” who look and behave like people, but they’re not people. Now imagine you can map someone’s likeness atop this technology.
Honestly, NEON wasn’t that impressive and quickly devolved to the uncanny valley — the concept suggests that humanoid objects which imperfectly resemble actual human beings provoke uncanny or strangely familiar feelings of eeriness and revulsion in observers).
The uncanny valley is why @lilmiquela just never seems quite right.
But you can see how this technology will continue to be foisted upon us.
It’s coming to mass entertainment. In fact, Disney Research and ETH Zurich teamed up on an algorithm for automatic face-swapping in photos and videos, which has a high enough resolution for filmmaking. It’s the first deepfake method with a megapixel resolution, which means it would be good enough for feature-length movies…
Recent Disney movies, such as Rogue One and Star Wars: The Rise of Skywalker have used face-swapping technology to have living actors perform the roles of other actors who have passed away with varying degrees of success.
The new method for face swapping, outlined by Disney, is so eerily accurate that it could be used in movies and TV and greatly improve the famous company’s ability to bring actors back from the dead in a realistic way.-Interesting Engineering, June 2020
Feasibly, Princess Leia Organa played by Carrie Fisher’s ghost isn’t the last we’ll see of Hollywood’s tricks. Imagine an entire film using Samuel L. Jackson’s deepfake that he doesn’t act in, while he’s still living. It could happen. Something like that will happen. And it will be clearly entertainment and mostly harmless.
POLICING THE DEEPFAKES
Despite all of these examples and the energy behind them, deepfake technology is not actually fun nor harmless. In an era of “fake news” and “alternative facts,” all that our society has left to determine if something did or did not happen is proof.
“Proof” that often takes the form of video, especially given the ease of doctoring printed images and photos. Video, for better and worse, is one of our culture’s only means of definitively backing up a fact.
And the window of using video as proof is quickly closing.
To help stop the spread of fake and doctored images, Google has released a tool called Assembler that looks for telltale algorithm tweaks in images to notify journalists if they have been faked.
Fake-spotting could end up being a huge business vertical, and look for these kinds of tools to start being incorporated into all social networks as a general rule. And they will need to encompass deepfakes, which will require a huge amount of primary data to cross reference against. Perhaps unrealistically.
You may have seen fake-spotting alerts on Instagram. When the algorithm indicates an image has been altered, it flags it, vets it, and then shares this notification. When this capability went live it started tagging memes, which are purposefully altered, to much chagrin.
But when we think of asking the social networks and news networks to fact check video of humans, there is a lot more processing power and data needed. Most of us have been accidentally tagged as someone else on Facebook. My Facebook always tries to tag my wife in pictures of my daughter, for example. Facebook requires a huge amount of first-party data to know who is who. Lots of pictures of my wife and daughter to improve accuracy. It’s not an easy solve.
Adobe, Twitter, and The New York Times Company are working together on a new system for adding attribution to photos and other content, called the The Content Authenticity Initiative (CAI), that could help align different orgs in policing deepfakes.
According to The Verge, “A tool will record who created a piece of content and whether it’s been modified by someone else, then let other people and platforms check that data.” Adobe debuted a prototype in Photoshop and this technology could be used proactively to help identify a photo’s source and ensure artists get credit for their work.
In August, the CAI published a white paper outlining a standard for content attribution that is fairly thorough. however, it ends in this way
…even an optimally designed system cannot ultimately succeed in a vacuum. We now begin the important work of deeper, more expansive collaboration with leaders in technology, media, academia, advocacy and other disciplines. With this first step towards an industry standard for digital content attribution, we look optimistically to a future with more trust and transparency in media.—The Content Authenticity Initiative – Setting the Standard for Digital Content Attribution
As noted above, memers are using this technology to make some hilarious content. But they are on the honor system of not abusing the technology or spreading “funny” content in a way others may believe it’s real. It’s not sustainable and realistic.
Without better identifying and policing, widespread adoption and misuse of deepfakes will impact society’s common understanding of things like facts, truth, and proof. Looking at the future optimistically just isn’t doing enough.
With this technology moving so quickly to mobile apps and faster processing, it’s difficult to predict that deepfake technology will not be heavily abused.
Of note, we are barely five years out from the most powerful elected representatives in our country bragging that they don’t use “get” email and that they still use flip phones.
Given the ambivalence of our legislators when it comes to 50 year-old email technology, let alone their clear misunderstanding of net neutrality, let alone their clear misunderstanding of social networks (“Senator, we run ads“), you will have to excuse my skepticism that they will be able to put firm guards around deepfake identification and policing anytime soon.
I will give the House Judiciary subcommittee on antitrust a little credit. But it’s not enough. Deepfakes are highly technical and their spread and deployment requires a deep understanding of machine learning, how content spreads in social network, privacy, moderation, parody rights, and more.
The current crop of elected officials and social network leaders aren’t prepared for this, I’m afraid. Especially in a world where many of those same leaders have purposefully disrupted the idea of “truth” and “facts,” our increased reliance on video to demonstrate proof is at-risk.
Here’s roughly what I predict may happen:
- Deepfake technology will continue to need less first-party data to create believable content (e.g., only need 1-3 photos or a few seconds of video)
- Mobile deepfake apps will continue to get faster and more accurate, which means greater accessibility and portability.
- Social networks will continue to use deepfake technology for fun experiences, filters, face-swaps and more. Advertisers will continue to use this technology. Memers will use them for jokes. This will continue to mask the seriousness of the deepfake threat.
- Education about deepfakes will continue in starts and spurts, but not really cross into the mainstream mind space on its own in any sort of way that will inoculate against its misuse.
- A deepfake scandal will rock social networks and news networks, possibly sparked by memers. Legislators and stakeholders will act shocked about this new technology and how they could never have seen its misuse coming. After a couple weeks of news cycles about it, we’ll move on. Repeat. Until something huge and culture-changing happens using deepfakes. By then, the technology will be so accessible and rampant, it will be too late.
Deepfakes are not inherently evil, but I believe they will be used for evil purposes. Hope I’m wrong. Until then, if you see a video of me with my head on a body looking just a little off, please assume it’s a deepfake. It’s just easier that way. -Greg