Truth and Deepfakes

 In Robert Stone's 1974 novel Dog Soldiers, the cynical failed writer John Converse describes how he has come to make a dishonest living by practicing the dark arts of his authorial trade. His boss, a former Trotskyist-turned-tabloid tycoon, he tells us, operates a press that prints knock-off versions of mainstream magazines. Inside the covers, these publications are full of lurid tales of crime, sex, and violence that are invented out of whole cloth. Converse's task as a "reporter" for these pseudo-magazines is to find photos and make up the bogus stories to go with them. For this purpose, he relies entirely on pictures of the deceased, cropped from other sources, because of a loophole in copyright law that ensures the dead have no right to protect their likenesses from misrepresentation. 

This example from an old novel suggests, perhaps, that visual "deepfakes" are not unique to the digital age. But the capacity of AI to generate plausible visages and scenery for episodes that never actually took place, and for people who never existed, has undoubtedly accelerated the problem. In recent weeks, news stories have warned of the speculative dangers of politicians using AI to generate photographic disinformation about their rivals. They featured a few early real-world examples: there were the invented photos of Trump's arrest, for instance, which were created as a means to underline the dangers of AI disinformation, but which were subsequently disseminated on social media without any disclaimer. There were bogus photos of Pope Francis in a puffy designer jacket. And so forth. 

Most of these news accounts, as I say, had the tone of a prophet warning against a potential danger still barely glimpsed on the horizon. Yet, scarcely a week later, we've already seen AI-generated deepfakes erupt into the political mainstream. In response to the launch of Joe Biden's 2024 political campaign, the Republican National Committee put out a YouTube ad featuring AI-generated images of the hypothetical dystopian future that would supposedly result from Biden's re-election. The ads came with a disclaimer in exceedingly fine print that the images were created using AI, but this could easily be ignored or cropped. The ad therefore surely crosses a new threshold in the mainstreaming of disinformation. We now have our first case of a national political party deliberately seeding deepfake images. 

John Converse's fictional job in 1974 (which no doubt had its real-world analogues) may have been every inch as cynical and deceptive as the generation of AI political deepfakes. But it was unmistakably more plodding and time-consuming. He was forced to comb through and crop the images of the verifiably deceased from other periodicals. Today, the same sort of hypothetical tabloid could create infinite images of made-up people, through aggregating the likenesses of innumerable real people. The human motives involved might be no worse than Converse's—greed, cynicism, or simply the need to put food on the table; but the scale on which the lies could be created would be much greater. 

Most of us would accept in principle that such a proliferation of false information is a problem for democracy. The notion that truth and reliable public decision-making are connected to one another is commonplace by now, but wishing to understand more deeply the risks involved, I read Walter Lippmann's Liberty and the News today in preparation for this post. Lippmann's brief essays are something of an ur-text for the notion that freedom and truth in the media need each other, and it remains timely in multiple respects—both for the warnings it utters about what we would now call disinformation, as well as its prophecies as to the increasing dysfunction of Congress and the consolidation of power in the executive branch to fill the void. Lippmann's book is indeed worth reading again, now that AI graphics introduce a new degree of uncertainty into the question of whether we can trust what we see in the news or our social media feeds. 

Lippmann's proposed solutions for the propagandistic and mendacious "yellow journalism" of his day have aged less well. They divide neatly into categories of those now considered so obvious that they have long since been implemented (such as instituting higher professional and ethical standards in the nation's journalism schools), and those that smack to us in a skeptical age of creeping authoritarian technocracy (such as proposed regulations and accountability courts for newspaper editors, which would hold them to a higher expectation of trustworthiness), and therefore seem to invite as many potential threats to truth as they would resolve. One is reminded of proposals for eliminating disinformation from social media today: surely a laudable goal—but one that also invites a recursive loop, since the people who are deciding what counts as disinformation and what does not must themselves be accountable to some standard of honesty. 

Leaving aside the suspicions of elitism that haunt Lippmann and the other Progressives of his era, from a contemporary viewpoint, it is unmistakable that his book offers a prescient warning—finely-attuned to the dilemmas of our era—about the dangers of false reporting and disinformation. He has given us mottoes that could be emblazoned on every contemporary masthead, such as: "There can be no liberty for a community which lacks the information by which to detect lies." (There, surely, is a warning for the age of AI deepfakes, if ever there was one!) And some of his comments read almost as if they were written with some dim inkling of a future Trump in mind: "[T]he demagogue," he writes, "whether of the Right or the Left, is, consciously or unconsciously an undetected liar." 

And then there are the passages in which a statement Lippmann plainly intended at the time as a grotesque hyperbole, for purposes of making a point, has lived on to foretell something all too close to subsequent reality. Lippmann writes in 1920, for example, of this ease with which people can lie and be believed, so long as their lies coincide with the xenophobic or jingoistic prejudices of their audience. "Nobody will punish me if I lie about Japan, for example"—he writes. "And if there should be hostilities with Japan, the more I lied  the more popular I should be. If I asserted that the Japanese secretly drank the blood of children [...] that the Japanese were really not a branch of the human race after all, I guarantee that most of the newspapers would print it eagerly." 

The decades after Lippmann wrote this, of course, would in fact witness the internment of Japanese Americans by the U.S. government, part of a massive wave of xenophobia and racism that followed the attack on Pearl Harbor (of the sort E.E. Cummings satirized in a memorable and justifiably vitriolic little poem, written entirely in a phonetic rendering of a drunken lout's racist anti-Japanese diatribe, "YgUDah"). Xenophobia against immigrants and foreign countries is obviously a large and growing part of U.S. politics today as well. And as for Lippmann's remark about a hypothetical blood libel, which he plainly intended as hyperbolic in order to make a point—well, actual accusations that one's political enemies are ingesting children's blood have become a standard feature of contemporary American politics, through the rise of the Q-Anon conspiracy theory. 

Plainly, mad lies and disinformation could come to influence U.S. politics in ways that even Lippmann could only contemplate in the form of absurd hypotheticals. The power of the lie is as undiminished in our day as it was in 1920, and if anything it may be even greater. And the proliferation of AI deepfakes, and their already-apparent use in mainstream politics, including by one of the nation's two leading national parties, may well exacerbate this problem even further. If there is any truth to Lippmann's argument, these events certainly doesn't bode well for the future of our democracy.