It Would Be Very Easy for the Government to Build a Computer-Generated Likeness of Donald Trump

By
Photo: Chip Somodevilla/Getty Images

Today, Donald J. Trump will take the oath of office and become the 45th president of the United States of America.

Despite his love of attention, Trump has indicated that he would prefer to conduct many of his presidential duties from Trump Tower, rather than spend too much time in the White House. Trump spent much of the late campaign season (when he thought he’d lose), and then the subsequent transition period (when he couldn’t believe he’d won) on his perch high above Manhattan, meeting only with his closest advisers. A quote circulating from the president-elect this week — “I mean my day one is gonna be Monday because I don’t want to be signing and get it mixed up with lots of celebration” — could be interpreted as Trump approaching the presidency as a nine-to-five job.

One way of interpreting these facts would be to say that Trump has shown great interest in occupying the most powerful office in the world, and little interest in fulfilling the associated duties. Back in July, when Trump selected Mike Pence as his running mate, the New York Times reported that John Kasich had originally been presented with the position. Donald Trump Jr. reportedly told Kasich that Trump’s vice-president would be in charge of domestic and foreign policy (so, everything), while Trump would focus on “making America great again.”

The way things are shaping up, a Trump presidency will be defined by largely unpopular orthodox Republican policy, repackaged under, and sold via, the somewhat more popular Trump brand name. Just as in his business, the most powerful tool in the Trump arsenal is his brand, the Trump name and gilded image. Which raises an admittedly absurd question: Does the Donald Trump presidency actually need Donald Trump, the human vessel?

Obviously it needs the Trump character — Mike Pence will not be able to command the attention of the press in the way that Trump clearly can. But thanks to deeply terrifying new technologies, Trump the human is less necessary than ever. It would be easier than you might think to create a digital facsimile of Donald Trump, one that could be deployed if Donald Trump became incapacitated — or inconvenient — in some way. Despite his insistence that he is “no puppet,” it would hardly be impossible to craft and deploy a digital puppet Donald Trump.

From a historical perspective, there are far less severe precedents for shielding the public from aspects that might affect their faith in the president’s ability to lead. When Woodrow Wilson was incapacitated by a stroke in 1919, his wife, Edith Wilson, fulfilled most of his executive duties until he left office, more than a year later. Most of the country was unaware at the time. The wheelchair-bound Franklin Delano Roosevelt requested that the press not take photos of him walking or transferring into his car, and if the Secret Service spotted anyone doing so, they had orders to interfere.

Ronald Reagan, who was elected to office at age 69, only slightly younger than Trump, has been the subject of much speculation. One of his sons, Ron, has alleged that Reagan began exhibiting symptoms of Alzheimer’s nearly a decade before he was officially diagnosed (other family members and Reagan’s doctor at the time dispute this). According to the Los Angeles Times, “Most high-level White House aides believed that President Reagan was so depressed, inept and inattentive in the wake of disclosures about the Iran-Contra scandal early in 1987 that the possibility of invoking the 25th Amendment to remove him from office was raised in a memo to Howard H. Baker Jr., then Reagan’s chief of staff.”

Historical precedent for masking the president’s condition only gets us half of the way there. We still need technology capable producing a digital Trump that’s lifelike enough. To find out whether this was possible, I emailed digital facial expert Mike Seymour from Sydney University, who co-authors the industry blog fxguide.com.

Most recently, effects artists at Industrial Light & Magic working on Rogue One: A Star Wars Story reconstructed the likeness of Peter Cushing by poring over hours of decades-old footage from A New Hope to get small details, such as the way his lips parted, exactly right. But reconstructing people from old material is not new. The otherwise forgotten 2004 film Sky Captain and the World of Tomorrow re-created Laurence Olivier, dead for 13 years at that point, using archival footage of the young actor. More than four years ago, the Tupac hologram made waves at Coachella, kick-starting a cottage industry of touring, deceased musicians. “There is nothing that the artists at ILM did to produce the digital humans in Rogue One that could not be done today to produce a clip of a digital President Trump,” Seymour asserted.

There are, similarly, hours of old footage of Trump, from his years as a New York socialite mugging for the camera to his late-career turn as a reality-television mogul on The Apprentice. Hours of footage that can be studied for mannerisms and movement patterns, facial tics, and so on and so forth. The easiest solution, according to Seymour, would be to composite another person’s mouth onto old footage of Trump, but that limits what can be accomplished.

Creating a 3-D facial scan of Trump wouldn’t be that hard. “Producing a lifelike high-resolution model is not quick, but there are many senior artists around the world who could do a remarkable job,” Seymour wrote, “and this could be aided by using photogrammetry that would digitally reconstruct a model from a suitable set of press photographs.”

“The situation gets worse, or better depending on your point of view, if we look at what is happening in the research labs,” he added. “Teams such as Pinscreen, and USC-ICT are producing remarkable models of a human face from a single quality still image. While not yet commercially available, in their labs, using complex machine learning, computer vision, and AI, they are producing a believable model of any face from one photo.”

But why go through all that trouble? It would be easy to convince someone like Trump, with his love of the camera, to willfully submit to a 3-D facial scan. Just tell him that it’s for posterity. That’s the same reasoning by which the Smithsonian was able to obtain a 3-D facial scan of Barack Obama last year. Seymour added that the file’s handlers are “clever enough to never post that data on the internet.”

Here’s a demo from March 2016 demonstrating how facial retargeting works, mapped onto the faces of leaders like Trump, former president George W. Bush, and … Vladimir Putin.

To make things easier, the president mostly gives speeches from behind a desk or podium, saving technicians the need to model his lower body or have the movements performed by a motion-capture actor. (An email sent to ILM seeking comment on the plausibility of this hypothetical — what it would take to construct a digital Trump — went unreturned.)

Then there’s the speech aspect. Trump’s vocal stylings are … certainly unique, but they can be broken down and analyzed, as some linguists already have. Using a large enough corpus of text, chatbots can be modeled using past speeches as a seed for future phrases. In November, Photoshop-maker Adobe demoed a new type of technology called VoCo, short for “voice conversion.” In the demonstration, the software user was able to rearrange words in a phrase to create new audio snippets, and to use the existing audio to fashion new words that were not present in the original sample. Get a few Trumpisms (“bigly,” “yuuge”), throw in some policy specifics, and Trump’s handlers could make the digital impostor say whatever they wanted.

Pulling all of this together into a believable package is the most difficult step in the process. “Our brains have developed specially to process and deal with faces. The tiniest facial muscle movement can denote major emotional differences,” Seymour told me. “Today, gaining a believable lip-synced animation is extremely hard and only a couple of places in the film world can reliably deliver.”

But that will change. “This technology is growing nearly daily and deep learning and neural nets are contributing not only to modeling and facial tracking, but soon to fine detail, high-frequency animation also,” Seymour added. “The actual computer rendering of the face has already become remarkably complex, producing digital skin that accurately represents the way light both reflects off skin and is partially absorbed and scattered.” In the meantime, popular effects programs are already available for licensing — products like Chaos Group’s V-Ray, Autodesk’s Arnold, and Pixar’s RenderMan can all be had for the right price.

And I haven’t even mentioned that it is still the case that many technological advancements are developed by the military, and then trickle down to private industry. If movie studios can do this much, then maybe we should ask what the government is capable of. After we put on our tinfoil hats, of course.

How expensive this initiative would be is difficult to say. (If they really want a low-budget way to use the Trump avatar, his handlers need only gain access to his Twitter account.) The effort to create digital humans is still young but accelerating at a rapid pace, and the size of the project determines the size of the budget. The government could have a few guys who are smart with limited resources or you could have a team of dozens burning million-dollar budgets like a major Hollywood studio. It all depends on the complexity of the project and the level of fidelity aspired to. It’s worth noting that even the largest Hollywood film budgets pale in comparison to federal defense spending.

The big caveat, however, is that we live in highly skeptical times. Conspiracy theorists on Reddit are ready to pounce on anything they see that doesn’t add up. Any small glitch could set off alarm bells (getting eyes right is apparently a big stopping point for clearing the uncanny valley). Any big problem — like if a digital Trump were to start glitching out like Milli Vanilli’s ill-fated MTV performance — could prove disastrous. The solution would simply be to have Trump avoid any sort of live appearance; instead only showing up on televisions and computer screens. Our new president has already expressed a complete and utter distaste for the news media; the last press conference that he held in 2016 was in late July. Otherwise, he was content to avoid anyone who might scrutinize him. A digitized-Trump presidency might be one of taped statements and few public appearances.

Thanks to a surge in mobile viewing, poor bitrates, low-quality streams, and other compromises on audiovisual fidelity, it would be easy to skirt any close analysis of media produced by the Executive branch. “In reality, given our diet of YouTube-user-generated content,” Seymour observed, “a poorly shot, grainy, or lower-resolution clip of a digital President Trump would be most likely more believable than a clip of the quality seen in Rogue One.” That film even considered a similar tactic; if the CGI Cushing facsimile had been too deep in the uncanny valley, they would have presented him as a fuzzy hologram transmission instead.

Is this scenario ridiculous? Clearly, yes — it is unimaginable that Trump would be sidelined and that members of the federal government would secretly use digital techniques to plant words in the mouth of a CGI Trump puppet in order to further their own agenda. But it is not impossible: The tools exist, and the subject appears willing to let others do his job so long as he gets the credit. This scenario won’t happen, it couldn’t possibly. We said the same thing about Trump winning the presidency.

It Would Be Very Easy to Build a CGI Donald Trump