How Archivists Could Stop Deepfakes From Rewriting HistoryOct 16, 2018
History is rife with fakes. In 1983, the German magazine Stern announced that it had acquired previously undocumented diaries written by Hitler, a find British historian Hugh Trevor-Roper initially heralded as "an archive of great historical significance." In reality, however, an illustrator named Konrad Kujau had penned the volumes himself. Thanks to the scrutiny of historians at the German Federal Archive, they were soon revealed to be forgeries and the so-called "Hitler Diaries" became a cautionary tale about media frenzies.
Imagine, however, if experts couldn't readily identify the dairies as fraudulent. And imagine, also, if forgers were able to create and distribute ultra-realistic fake Nazi records at breakneck speed. Finally, imagine if some of these documents were forever preserved as authentic pieces of Nazi history. A threat like this is edging increasingly out of the hypothetical and into the real as the tools to quickly create realistic manipulated videos go mainstream. These videos, which use machine learning to graft one person's face onto the body of another, are known as deepfakes--and they're getting disturbingly good.
While many have feared the potential of deepfakes to spread misinformation in the here and now, these videos could distort reality long after today's fake news goes viral if they're improperly archived as legitimate. Gizmodo spoke with several historians, archivists, and professors who were familiar with the deepfakes phenomenon, some of whom had pragmatic concerns about it. Fortunately, archivists have rigidly established principles meant to catch forgeries and screw-ups, but these protections are only as strong as the institutions that provide them.
Roger Macdonald, the director of the television archive at the Internet Archive, characterized deepfakes as "a looming threat" since last year, when researchers showed they could create realistic fake videos of former president Obama synced to audio clips.
"When the University of Washington team put out the synthetic Obama, we realized the public threshold had been crossed," said Macdonald, "and we share the concerns of other archivists and lots of folks within media and otherwise who care about reality."
David Staley, an associate professor at Ohio State University who teaches courses in digital history and historical methods, echoed Macdonald's sentiment. "As a historian, one of the things I'm concerned about is what happens to the record that is left behind if a video is doctored, for instance, and it's not readily apparent that it has then been" said Staley. "That has all kinds of implications I think."
While deepfakes pose a disturbingly high-tech challenge to fact-checkers, victims, and the general public, however, disinformation itself is hardly novel. For at least one archivist, these videos are just the latest version of age-old problems like fraud, manipulated media, and propaganda.
"It's something archives have dealt with for centuries," Yvonne Ng, a senior archivist at WITNESS, a nonprofit that focuses on collecting video evidence of human rights abuses, told Gizmodo. "The deepfake is a new spin on this process, but archives have always had to deal with forgeries or fakes or plagiarism--and even unintended damage and deterioration--and then having to determine the authenticity of objects with all of those considerations in mind."
Archivists have been concerned with provenance, or the origin and chain of custody of information, well before the emergence of deepfakes and digital manipulation tools. Ng said that a good historical analogy to deepfakes for archivists would be portrait paintings, noting that these were often painted long after their subjects had died. She said that you can still document such a painting in your archivist collection, but there would be different things to examine to determine just how truthful the work is. She said you could look at the clothes and buildings in the painting in order to see if they are representative of the period of time when the subject was alive, as well as who the painter was, and if they were alive at the same time as the subject. At a physical level, an archivist might look at the format of the painting, what type of paint and materials were used, and if those could be dated, Ng said.
"Archival methods are not primarily about tools and the tech," Ng noted. "Archival methods have always been more about having controlled and consistent policies and rules." She said that descriptions of information and its metadata are a major part of archival work: In other words, documenting the context of the content.
Macdonald said that the problem isn't deepfakes entering the historical record entirely. In fact, he argued, they should be preserved. The videos are deeply representative of a time when disinformation is troublingly realistic and questions of authenticity are a public concern--a time when our ability to discern the truth is racing against technological advances.
"We don't fear documenting something," said Macdonald, "as long as we're being accurate about that which we document." He detailed a hypothetical situation in which a news program broadcast a deepfake video, whether or not they were aware that the video was manipulated. "We are really interested in making sure we preserve that," he said, "and so the only way that libraries and archivists can be challenged in terms of being fooled is having our understanding of the source be wrong or later corrupted."
The more disturbing possibility with deepfakes is that historical records will include manipulated media presented as reality, whether that's because an archivist wasn't able to identify a video's origin correctly or because someone was able to infiltrate records and manipulate them.
"The classic dystopian concern is represented in 1984, where history was destroyed," said Macdonald. According to Macdonald, the destruction of history is a major concern for the Internet Archive and similar projects, and the best way to prevent that is to maintain and protect a lot of copies. "It is vital that there be serious redundancy, and there is not," he said. "Secondly in the 1984 scenario, not only was history destroyed, but it was also rewritten, so it is vital that a variety of robust techniques are used to assure that the records that libraries hold have not been altered."
Nicolas Taylor, a program manager for the LOCKSS and Web Archiving programs at Stanford University, said that they "endorse a particularly paranoid approach to digital preservation." LOCKSS stands for "lots of copies keep stuff safe," and is the principle guiding Stanford's program to afford libraries and publishers with the resources they need to digitally preserve content forever.He said that attacks on their systems--whether by an outside hacker or a privileged insider with access--are something they take into consideration. And it wouldn't have to be someone with malicious intent: they could have simply made a mistake.
"We want to build systems where tampering is evident and it prompts alerts," said Taylor. "We want to build a system where the integrity of the information is determined by the consensus of peers."
If appropriate redundancy measures are taken, Taylor suggested that an attacker would have to compromise all copies of the information nearly simultaneously. This would take some impressive coordination, and as troubling as it is to imagine a future where digital propaganda is preserved as official history, deepfake attacks are far from a priority of archivists.
Taylor said that memory institutions and digital preservation projects are mostly worried about "benign and passive threats to the integrity of digital information over time," such as "bit rot" (a term for the gradual decay of storage data), the failure of storage media, environmental threats, natural disasters, and hardware and software obsolescence.
"I think there's an interesting question about if the value and purpose of deepfakes is to serve some propagandistic aims," Taylor said. "It may only be sort of a particular subset of deepfakes where it's really valuable to try and launder them through an archive rather than try to see their maximum effect through more mainstream channels."
Gregory Wiedeman, a university archivist at University at Albany, SUNY, said deepfakes are "more of a long term issue." "It sounds a little scary and we have more immediate goals that I'm more concerned about," said Wiedeman, "but I also think we have sort of procedures and best practices in place and we hope we can rely on them. And if we find out we can't, we make adjustments."
For archivists, it's understandably difficult to dwell on a potential future threat when you're still keeping up with the past and present. And the experts I spoke to seemed impressively well-equipped to tackle the challenge of deepfakes and to ensure that they are documented correctly and redundantly.But even if history isn't recorded misguidedly or maliciously rewritten, future generations will need to equip themselves with the ability to perceive what's real and what's not.
The story of the Hitler Diaries ends with an instructive coda. In 2013, many of the forged diaries were donated to Germany's Federal Archive--the very institution that had exposed them as a fraud decades earlier--as part of journalism history. "These documents are of great significance to past history and the history of the press," said an archive spokesperson at the time. "Everything you read in the papers that is cold coffee by tomorrow, we preserve for eternity."
Ultimately, the greatest protection archives offer against the distortion of history may be their careful documentation of previous errors. By supporting archiving projects, we not only ensure that the past is preserved accurately, but create a guide for the future by chronicling the long relationship between media and deception.
Macdonald argued that archivists, libraries, and society in general needed to "invest in its memory," saying, "I think it's very important that lots of folks become aware that our record of what happens could be destroyed or could be altered."
Today, that threat to our collective memory might be bit rot. Tomorrow, attacks by deepfake propagandists. The day after that, some currently unimaginable danger. In the meantime, we'd be wise to not only keep lots of copies, but do our best to look at them regularly and try to learn from our mistakes.