The Danger of Deepfakes

A recent article on the Forbes website shows photos of four fairly normal-looking faces:  two young women, a more mature woman, and a man.  Thing is, none of these people exist.  The faces are composites made with artificial-intelligence (AI) technology that produces what are called deepfakes:  fictitious images that only sophisticated specialists armed with software can tell from the real thing, and sometimes they can't even tell either. 

Deepfakes aren't limited to still pictures.  Another example the article cited was a State Farm commercial that purported to show a 1998 ESPN commentator making eerily accurate predictions about events in 2020.  There was no such commentator, but deepfake technology enabled the producers to make it look that way.

I've blogged about deepfakes before, but in our current volatile circumstances, the topic deserves some revisiting.  Like almost anything having to do with AI these days, the quality of deepfakes is increasing as the computer horsepower needed to make them decreases.  One way AI developers have found to improve the quality of deepfakes is to pit one program against another in what's called a "generational adversarial network," or GAN.  It's a digital version of what a writer does when she first throws down anything she can think of on a page, and then takes off her creative cap, puts on her editor's cap, and looks for the best parts of what she's written and develops those.  One system comes up with attempts at deepfakes and the other system critiques them, and together the two systems can approach something closer to realistic images than either one can by itself.

Regardless of how it's done, it's becoming harder to tell a fake still or video image from the genuine article, and thereby hangs the rub.

Every so often, the body politic enters what one might call a critical moment.  We've had more than our share of these lately.  I'd say one critical moment came last March over the weekend of the 13th to the 16th.  That was when the full implications of the COVID-19 pandemic registered with officialdom, and emergency-restriction edicts began to show up everywhere from the local to the national level. 

Another critical moment came after news reports of what happened to George Floyd in Minneapolis on May 25 included the famous nine-minute video of officer Derek Chauvin pinning Floyd's neck under his knee.  In today's video-saturated age, no number and quality of words can have the effectiveness with large masses of people that a single video clip can have. 

There was never any serious question about the video's authenticity, which added to its impact.  Without getting into the complicated issues of the relationship between videos and total reality, I will simply point out that the Floyd video created a hypersensitized public ready to be outraged by anything similar in nature, whether real or otherwise.

There are actors and institutions out there which would like nothing better than to sow discord in a city, a state, or a nation.  The 2016 presidential elections provided abundant evidence that Russia engaged in a well-resourced attempt to influence the election in ways that were disguised to appear as simply concerned U. S. citizens expressing their opinions, or spreading rumors. 

We would like to believe that truth will eventually triumph, that lies and fakery eventually get discovered and discredited, and that if you just explain the facts to people in a logical way that they will agree with you.  But in the heat of the moment, such as the runup to an election, a lie can get five miles down the road while the truth is still getting its pants on, and that can change the course of history in an election.

Most of the deepfakes that have drawn much attention so far are extremely unbelievable, and made so intentionally to show off their apparent authenticity.  The ESPN prophet is one example.  When a deepfake specialist makes former President Obama say something that no one would believe he'd actually say, the result isn't going to fool anybody.

For a deepfake to cause serious trouble, it would require a coordinated effort among the deepfakers and a conspiracy of false witnesses to the alleged event.  Just to play devil's advocate for a moment, you may recall some time back when actor Jussie Smollett had two people attack him in Chicago with slurs and a rope around his neck.  I chose the construction "had two people" intentionally, because a police investigation subsequently turned up strong evidence that Smollett paid his attackers to stage the incident for publicity purposes. 

What if Smollett had instead hired a deepfaker to create a video of the incident, along with suborning witnesses who would testify to its authenticity?  Now you've got potential dynamite instead of just a little firecracker that would take authorities just a few days to see through.  And the ever-roving eye of the public, greedy for outrageous confirmation of its suspicions, would eagerly seize upon such evidence of bigotry and go to town, so to speak.  Smollett would have stood a much better chance of being believed, at least for a while, and it's possible that lots of people would be convinced by the video and might not ever change their minds afterwards.

The Smollett incident happened in January 2019, which in comparison to today seems like the bygone Edwardian era of England before the catastrophe of World War I.  If  someone fabricates a deepfake-assisted fraudulent incident that is primed to touch a sensitive nerve such as racism or the intense dislike that members of the public have toward a prominent political figure, it will not take much to set off reactions such as we have been seeing for the last couple of weeks since George Floyd's death. 

Over the centuries, society has learned how to cope with forgery in documents and paintings, counterfeiting in currency, and darkroom manipulation of still photographs.  As the technology of deepfakes advances, I expect that we will also learn how to deal with the fact that no matter how realistic a video looks, there is always the possibility that it was faked.  And that thought is especially important to hang onto in distressed times such as these, and in critical times leading up to important elections.

Sources:  The Forbes article by Rob Toews appeared on May 25, 2020 at https://www.forbes.com/sites/robtoews/2020/05/25/deepfakes-are-going-to-wreak-havoc-on-society-we-are-not-prepared/.  I also referred to Wikipedia articles on Jussie Smollett and George Floyd.  I thank my wife Pam for bringing this article to my attention. 

Seeing Hasn't Been Believing For Some Time

As you may have been reminded recently, the second person to walk on the moon was Edwin Eugene "Buzz" Aldrin Jr., who accompanied Neil Armstrong in the lunar module during the Apollo 11 flight in July of 1969.  A common problem faced by the lunar astronauts was what to do with the rest of your life afterwards, and one thing Aldrin did was to make public appearances about his astronaut career.  On Sept. 9, 2002, Aldrin showed up at a Beverly Hills hotel expecting to be interviewed on camera for a Japanese children's television program.  Instead, waiting for him was one Bart Sibrel, a sometime documentary filmmaker who has made a career out of promoting the idea that NASA's moon landings were all faked in a secret CIA-operated studio.  Sibrel, accompanied by his own film crew, aggressively tried to get Aldrin to swear on a Bible that he had landed on the moon, and when Aldrin declined to do that in front of cameras and told Sibrel to leave him alone, Sibrel called him a "thief, liar, and coward" and reportedly backed him against a wall and poked him with the Bible.  In response, the 72-year-old Aldrin punched the 250-pound Sibrel in the jaw, but no charges were filed. 

Sibrel has made a career out of claiming that all the visual evidence (as well as physical evidence in the form of moon rocks distributed around the world) showing that men have been to the moon was essentially an elaborate "deepfake."  The word was not invented in 1969, but the concept of lying is as old as humanity.  For reasons of his own, Sibrel thinks (or at least appears to think) that NASA and the CIA concocted an extremely elaborate lie and backed it up with artificially-generated visual evidence. 

In 1969, there was no such thing as advanced computer graphics that could take images from different sources and combine them seamlessly to make it look like, for example, actor Tom Hanks was in the same room with President John F. Kennedy, as the movie "Forrest Gump" showed in 1994 in one of the first films that took advantage of computer-generated imagery (CGI).  Well, the great democratizing force known as IT has now brought the simpler kinds of digital fakery to the masses.  Some people are worried that simple tricks such as speeding up or slowing down authentic videos will cause more trouble than the sophisticated deepfakes that even experts have problems detecting. 

A recent Associated Press article describes how U. S. Speaker of the House Nancy Pelosi was made to appear physically impaired by simply playing back a recording of her at a slower speed.  While some such tricks are pulled purely for satirical purposes, digital forensics expert Hany Field worries that unsophisticated voters will be fooled by them nonetheless.  He called the doctored Pelosi video, which got over two million views on Facebook, "a canary in a coal mine," and expects the 2020 election year will see many more such crude trick fakes, which one could call shallowfakes.

According to some studies quoted in the article, some groups of voters (older ones and "ultra-conservatives," whatever that means) tend to trust videos more and will retweet, and otherwise treat as credible, videos that younger and less conservative people will quickly recognize as having been altered.  I have seen this sort of effect myself as I have watched an otherwise sensible and well-balanced woman of my acquaintance lap up stuff on Facebook that I consider arrant nonsense, on occasion. 

It's possible that such people hold all Facebook information in a different category in their minds from things that people they trust tell them in person, and that it's a sort of entertainment more than a serious search for the real truth.  But all this lies in the uncertain realm of what may influence voters, which includes everything from the weather to the state of one's health.  And as such, we can only speculate on its effects.

Two possible ways mentioned by the news article to mitigate any negative effects of shallowfakes are so-called "downranking" by social media outlets such as Facebook, and adding fact-checking information to suspicious-looking videos or information such as labels saying "This video has been altered from its original state."  The problem with this is that it puts the social-media operators in the role of censor, or at least editor.  And depending on the political character of the shallowfake and the political leanings of the censor or editor, someone is sure to cry "Foul!" at such actions, or at least point out equally egregious shallowfakes on the opposite end of the political spectrum that haven't been censored, edited, or labeled as such.  That is a quagmire that organizations like Facebook may not care to wade into, and I'm not sure they should.

One thing is for certain:  the video-altering genie of cheap and easily available software is not going back into its bottle.  And any thought of government regulation or censorship moves us in the direction of dictatorships like China, where members of minority groups such as Uighurs get sent to what amounts to concentration camps simply for trying to get in touch with other Uighurs on social media. 

So perhaps the answer is a better-educated electorate.  Civics education in this country is reportedly in terrible shape anyway, as many history texts seem to concentrate on everything bad our forefathers did (excuse me, should I say "forepeople"?  No, I shouldn't.).  I do recall one lesson I learned from a high-school history book, which was that after the famous purges of the 1930s in the Soviet Union where whole rafts of bureaucrats were executed at Stalin's whim, new editions of history books carried photos of the Great Leader from which certain undesirables had been airbrushed away, without comment.  I was warned in the 1970s against that sort of thing happening, and so today we should warn our high-schoolers against the danger of both deepfakes and shallowfakes.  But after that, it's up to them to judge—and to vote.

Sources:  The Associated Press article about deepfakes and shallowfakes by Beatrice Dupuy and Barbara Ortutay, dated July 19,was carried by many news outlets, including the Longview (Texas) News-Journal at https://www.news-journal.com/ap/politics/deepfake-videos-pose-a-threat-but-dumbfakes-may-be-worse/article_20f82829-5abd-5e83-8456-2630fc4b097b.html.  The Aldrin-Sibrel incident is described in the Wikipedia article "Buzz Aldrin."

Seeing May Not Be Believing: AI Deepfakes and Trust in Media

The 1994 movie “Forrest Gump” featured authentic-looking newsreel footage of the 1960s in which President Kennedy allegedly appeared with Tom Hanks’ fictional Gump character.  Trick photography is as old as cinema, and the only remarkable thing about such scenes were the technical care with which they were produced.  At the time, these effects were state-of-the-art and took substantial resources of a major studio.

What only Hollywood could do back in the 1990s is soon coming to the average hacker everywhere, thanks to advanced artificial-intelligence (AI) deep-learning algorithms that have recently made it possible to create extremely realistic-looking audio and video clips that are, basically, lies.  An article in October’s Scientific American describes both the progress that AI experts have made in reducing the amount of labor and expertise needed to create fake recordings, and the implications that wider availability of such technology poses for the already eroded public trust in media.  Fakes made with advanced deep-learning AI (called “deepfakes”) can be so good that even people who personally know the subject in question—President Obama, in one example—couldn’t tell it was fake.

The critical issue posed in the article is “. . . what will happen if a deepfake with significant social or political implications goes viral”?  Such fakes could be especially harmful if released just before a major election.  It takes time and expertise to determine whether a video or audio record has been faked, and as technology progresses, that difficulty will only increase.  By the time a faked video that influences an election has been revealed as a fake, the election could be over.  We faced something similar to this in 2016, as it has been conclusively shown that Russian-based hackers spread disinformation of many kinds during the runup to the Presidential election. 

Some voters will believe anything they see, especially if it fits in with their prejudices.  But the firmer a voter is embedded in one camp or the other, the less likely they are to change their vote based on a single fake video.  The people who can actually change the outcome of an election are those who are undecided going into the final stretch of the campaign.  If they are taken in by a fake video, then real harm has been done to the process.

On the other hand, the public as a collective body is not always as stupid as experts naively think.  If a deepfake ever manages to be widely believed at a critical moment, and the fakery is later revealed publicly, the more thoughtful among us will will keep in mind the possibility of fakery whenever we watch a video or listen to audio in the future.  This can be likened to an immune-system response.  The first invasion of a new pathogen into one’s body may do considerable damage, but a healthy immune system creates antibodies that fight off both the current infection and also any future attempts at invasion by the same pathogen.

If deepfakes begin to affect the public conversation significantly, we will all get used to the fact that any audio or video, no matter how genuine-looking, could be the concoction of some hacker’s imagination. 

Low-tech versions of this sort of thing happen all the time, but with lower stakes.  When I’m not writing this blog, I find time to do some lightning research, and a few years ago someone forwarded me a YouTube clip purporting to be a security-camera video of a guy who got struck by lightning, not once, but twice, and survived both times.  I watched the grainy monochrome recording of a man walking toward the camera on a sidewalk.  Suddenly there was a bright full-screen flash, and he’s down on the pavement, apparently dead.  Then he raises his head, shakes himself, and groggily rises to his feet, only to have a second flash knock him down again.  I heard from another lightning expert about this video that it was definitely fake.  Some people want so desperately to achieve viral fame, that they will go to the trouble of setting up an elaborate fraud like this one just on the hopes that their production will be kooky enough to get shared widely.  And in this case, they succeeded.

Speaking theologically for a change, some (including myself) trace the origin of lies back to the father of lies himself, the devil, and attribute lying to the only Christian doctrine for which there is abundant empirical evidence:  original sin.  No amount of high-tech defense is going to stop some people from lying, and if they can bend deep-learning AI to nefarious purposes such as creating inflammatory deepfake videos, they will.  The best defense from such scurrilous behavior is not necessarily just working harder to make fake-video-detection technology better, although that is a good thing.  It is to bear in mind that people will lie sometimes, and to use the time-honored rules of evidence to seek the truth in any situation.  And to bear in mind something that is often forgotten these days, that there is such a thing as objective truth. 

I think a more serious problem than deepfake videos is the fact that in pursuit of the online dollar, social media companies have trained millions of their customers to react to online information with their lizard brains, going for the thing that is most titillating and most conforming to one’s existing prejudices regardless of the likelihood that it’s true.  They have created an invisible mob eager and willing to be set off like a dry forest at the touch of a match.  And once the forest fire is going, it doesn’t matter if the match was real or fake. 

Sources:  Brooke Borel’s article “Clicks, Lies, and Videotape” appeared on pp. 38-43 of the October 2018 issue of Scientific American.