by Joseph Shieber
1. Bored, and with little to occupy their time, two cousins, Elsie, who was 16, and Frances, who was 10, decided to play around with photography. At a river near where they lived, they manipulated an image so that it looked as if they were interacting with little, magical winged creatures — fairies.
The photo was believable enough that they fooled a number of adults — including world-famous writers. The girls produced a number of other photos, using the same methods. The media was ablaze with discussions of the images and of whether they provided proof of the existence of fairies.
This all happened in 1917.
I was reminded of this case — the case of the Cottingley fairies — by the recent interest in the phenomenon of deepfakes.
Deepfakes are incredibly realistic manipulations of video and audio. Here, for example, is a video of President Obama uttering something that President Obama never said — made by swapping in the actor Jordan Peele’s mouth and voice.
If you believe the hype surrounding deepfakes, this technology threatens not only “the collapse of reality”, but also the falsification of our memories. While the threat is real, the problem isn’t actually with the deepfakes — it’s with us.
Actually, the discussion of deepfakes can help us to see two different problems that we face. Solving those problems, however, doesn’t really involve technological solutions.
2. Here’s the first problem: we’re not good at handling complexity. We want simple, easy-to-understand, answers. “1+1=2”, rather than “P(A|B) = (P(B|A) x P(A))/P(B)”.
The unstated premise in the widespread panic over the rise of deepfakes is that, prior to this technology, video evidence was supposed to be the gold standard for evidence, practically guaranteeing the truth of what it depicted.
This, however, is absurd. There is no gold standard, no single piece of evidence that can guarantee the truth of the information it conveys.
Let me be clear what I mean by that.
By saying that there is no single piece of evidence that can guarantee the truth of the information it conveys, I’m not one of those post-truth types who is trying to argue that there is no such thing as truth — nor, even, that there is just no such thing as truth anymore.
There is truth. Black holes exist. So does human-caused climate change. Vaccines don’t cause autism. John Lennon didn’t really “bury Paul” McCartney.
Instead, what I’m saying is that we don’t discover the truth by means of a single, glaring piece of evidence that removes all doubt. On the contrary, we only get at any truth — or at least any truth worth arguing about — by gathering and weighing evidence.
3. In the case of deepfakes in particular, there seem to be two difficulties with the idea that video evidence was ever the sort of knockdown, infallible evidence that the moral panic over deepfakes suggests.
The first difficulty is that, even prior to the existence of deep fakes, it was already possible to manipulate events so that they suggested an underlying narrative that wasn’t accurate. Here are two examples, one drawn definitively from the realm of conspiracy theory, the other … well, who knows?
Here’s the first example. Some people believe that Lee Harvey Oswald was a patsy, set up to be a fall-guy as the “lone gunman” who assassinated John F. Kennedy. Many of those who argue in that way aren’t trying to suggest that Oswald wasn’t in the Dallas Book Depository or that he didn’t have a sniper rifle or that he didn’t fire it at Kennedy’s motorcade. Rather, they’re saying that Oswald didn’t succeed in killing Kennedy, that there were other, hidden gunmen who actually committed the assassination, leaving Oswald to take the blame and to provide a convenient story to deflect a more comprehensive investigation.
The Oswald example, of course, is the stuff of conspiracy theories. The fact that the case is even possible, however, suggests that the threat that seemingly dispositive evidence might not offer an ironclad guarantee is not a threat that is new to the deepfake era.
So you don’t like evidence drawn from tinfoil-hat level conspiracy theories. I get that. Here’s a case that came to life more recently that might seem a bit more plausible.
In a recent piece for the Atlantic, James Fallows suggests that the evidence that brought down Gary Hart as a serious contender for the 1988 presidential election might have been a set-up initiated by Lee Atwater.
To anyone who didn’t follow politics in the 1980s, the names “Gary Hart” and “Lee Atwater” might not mean very much. But if you did follow politics in the 1980s, they mean a great deal.
Hart was everything that many Democratic voters thought they needed after the debacle of Reagan’s obliteration of Mondale in the 1984 election. Hart was young. He was from the West (Colorado, to be exact). He was strong on national security. He was brainy. He was handsome. In short, many thought he would be a very serious challenger to George H. W. Bush if the then-Vice President Bush decided to run on the Republican ticket in 1988.
Hart also had a reputation as a womanizer.
Here’s where Lee Atwater comes in. Atwater was responsible for many of the dirtiest of dirty tricks that led to Republican victories in the 1970s and ‘80s, including — perhaps most famously — the “Willie Horton” attack ad. That ad, against the person who actually ran against George H. W. Bush in 1988, Michael Dukakis, was an overt smear attack on Dukakis as being “soft on crime”. Implicitly, however, the ad was a racist dog-whistle intended to tie the image of Dukakis, a bespectacled former professor who had served as Governor of Massachusetts, to that of Horton, whose picture is a mugshot of a menacing African American man found guilty of multiple murders.
Atwater died of a brain tumor in 1991. Before his death, he repented all of the smears that he had perpetrated on his political opponents. For example, he called Michael Dukakis and apologized for the Horton ad.
And, as it now turns out, he also called a high-level Democratic operative who was working on a possible 1988 Hart presidential campaign to apologize for the incident that eventually caused Hart to bow out of the race.
Here’s the short version of that incident. Hart was photographed with a bikini-clad woman named Donna Rice sitting on his lap, aboard a sailboat with the suggestive name “Monkey Business”. Some weeks later, Hart and Rice were photographed entering Hart’s Washington, D.C. townhouse. After the photos emerged and led to persistent questions about Hart’s marital fidelity, Hart left the race.
According to Atwater’s confession, many of the elements of that incident: that Rice was on the sailboat with Hart, that she happened to be sitting on his lap at the precise moment the photo was taken, even the name of the sailboat — “Monkey Business” — all of those elements were orchestrated by Atwater.
That isn’t to say that the events didn’t occur; they did. Nor is it to say that the photos were faked; they weren’t. It is to say that all of the aspects of the incident that made it catnip for the Washington press at the time — who fools around on their spouse on a boat NAMED “Monkey Business”, for goodness’ sake?! — were the result of Atwater’s machinations.
What this demonstrates is that, even before it was possible to falsify photo or video evidence convincingly, it was already possible to manipulate events so that the photo or video evidence would seem to fit a narrative that served the manipulator’s purpose.
4. That’s the first difficulty with the current deepfakes panic and our naive expectation of overly simplistic explanations. It has always been possible to manipulate events so that they suggested an underlying narrative that wasn’t accurate. Here’s the second difficulty.
It’s one that is easy to appreciate if you’ve ever seen virtually any theater — whether high tragedy or low comedy. It’s at the root of tropes like “It’s not what it looks like!” More generally, it’s the problem of misunderstanding through a failure to appreciate context.
Here’s a made-up example of what I mean. Consider a case in which two soldiers are trapped behind enemy lines. The enemy is particularly cruel and vicious; any soldiers that they capture will be tortured before being killed. One of the soldiers becomes so badly injured that there is no chance of survival, and is too heavy for the other soldier to carry back to their own troops. Knowing about the enemy’s reputation for cruelty, the injured soldier begs to be killed and, in order to prevent the injured soldier from falling into enemy hands … and certain torture, the uninjured soldier agrees to perform a mercy killing. Unbeknownst to the two of them, however, someone has taken a photo of the moment at which the uninjured soldier fires the fatal shot to put the injured soldier beyond the reach of the enemy.
If someone later were simply to see the photo, they would see one soldier killing a comrade soldier, seemingly in cold blood. Yet this is a case that would have involved no manipulation — neither manipulation of the image itself nor of the events leading up to the fatal shot.
This difficulty, in other words, is distinct from the first difficulty because it doesn’t involve outside manipulation. It’s enough that life is complex. Situations often arise that can easily be misinterpreted if you lack the proper context. And the problem is that any individual representation of that situation will be limited. It will leave out some of the context that might well be needed accurately to understand the situation: either the before-and-after, or the world beyond the scope of the lens and out of range of the microphones.
5. In other words, the first problem posed by the phenomenon of deepfakes is actually a more general problem that we face whenever we have to evaluate evidence — even evidence that hasn’t been faked. It’s that we’re generally bad at dealing with complexity, context, and subtlety.
The second problem is that almost all solutions discussed for dealing with the first problem involve individualistic approaches. Almost universally, writers on the problem of deepfakes have suggested that the challenge of new media technologies will require greater analytical skills on the part of media consumers. (For a recent example, see Gerald Dworkin’s piece in 3QD here.)
This suggestion that appealing to individual consumers of information to be more vigilant, more analytic, and more rational is utterly bizarre but almost omnipresent. Here’s why it’s bizarre. The suggestion is that the solution to the problem that humans are bad at individual rationality is simply to implore humans to be better at individual rationality! It’s about as effective as asking the kids at Chuck-E-Cheese to eat more salad while piling the table with ever more and different varieties of pizza and cupcakes.
Many now recognize that the answers to systemic problems like gun violence, climate change, and the spread of infectious diseases cannot be found simply in appeals to individuals’ virtues, but will involve systemic and institutional approaches. Surprisingly, many of those same thinkers fail to acknowledge that the problem of misinformation and distrust of the media is also a systemic problem. Like those others, it also demands systemic and institutional solutions.
6. If the solution to the problem of media manipulation and misinformation can’t be found in individualistic appeals to — and unrealistic demands on — the rationality of lone media consumers, where might a solution be found?
The answer that I favor involves socially distributed cognition — collective information-seekers and decision-makers whose abilities to filter out inaccurate information exceed those of individual thinkers. In other words, the answer involves the sorts of systems and institutions studied by the nascent field of social epistemology, among others.
7. One of the worrisome aspects of the socially distributed approach that I favor is that it doesn’t provide any knock-down arguments against someone who would challenge me that the social systems in which I’m embedded — the networks of news, science, and academics from which I derive my information — are in fact reliable sources of information.
The problem is that accurately assessing the effects of systems and institutions is messy. Think about analyses of gun violence or disease spread or climate change: although there are often accounts and explanations of such phenomena that are widely accepted, there will always be room in the data for alternate explanations, no matter how fringe.
Analogously, the devotee of the Drudge Report or Fox News or QAnon could question my reliance on news sources like The Guardian, the Washington Post, The New York Times, and the Frankfurter Allgemeine Zeitung. I can explain that reliance, but of course it won’t be an explanation that will satisfy the committed Fox News viewer.
At most one of us is relying on accurate sources; neither of us has an account of that reliance that will convince the other.
8. Returning to the problem of deepfakes, my preferred solution is this. We — people like me, who prefer NPR to Breitbart News — need to strengthen the sorts of media and other institutions that have the resources to find out which evidence is reliable and which is fake. We need to influence discourse by expressing our preference for conducting discussion and debate on the basis of that sort of evidence. We need to elect public officials — beginning at the local level with school boards and town councils — who share our commitment to evidence-based decision-making, in which the evidence is drawn from the sorts of sources in which we have confidence.
In short, we need to work to shore up and invest in institutions that put a priority on accuracy and truth. (And yes, I write that with the understanding that those who disagree with me will have their own understandings of what the words “accuracy” and “truth” mean. Again: at most one of us will be correct. I’m betting on me.)
9. There is some connection between this discussion and a pair of posts that recently drew attention on 3QD (see here and here). The difference, though, is that the discussion of QAnon and skepticism in those posts largely involved evaluations of individual rationality.
If my discussion here is apt, however, the better diagnosis would be at the level of collective rationality. On that diagnosis, the correct response to QAnon isn’t along the lines of engagement and rational argument; rather, the strategy should be more along the lines of “Don’t feed the trolls”.