Abstract
As you may not know unless you have encountered Michael Grothaus’s book, deepfakes use artificial intelligence to map one person’s face or voice on to another, so convincingly that the resulting fake is impossible for humans to distinguish from reality. And on the dust jacket of Trust No One, the author outlines three scenarios illustrating how AI-powered digital fakery augurs a world where anything false can be made to look true. The doctored images show actor Margot Robbie having group sex, Xi Jinping declaring a nuclear war, and Michael Jordan winning the World Cup. But if these are real cases, they are not revisited in the book. A Google search returns no news reports of any such deepfakes having been created, let alone deployed to successfully deceive. Like almost all of the other examples of deepfakes presented by Grothaus’s book, they are probably hypothetical.
The question “what if?” suffuses the book. What if an election was swung by a deepfake? What if you were charged with a crime on the basis of a deepfake? What if someone created a deepfake of you and sent it to your friends, family and employers? These are legitimate fears, but Grothaus finds scant evidence of such frauds being perpetrated. The book opens with a vision of Joe Biden losing the 2020 presidential election as a result of a deepfake, which, as we know, did not happen. Two scenarios that he describes in detail – a deepfake of himself committing a crime, and another of his deceased father – are commissioned by him. While interesting, they clearly do not constitute real-world malice.
Frustratingly few voices of those closest to the issue can be heard here either. Legislators, copyright holders and privacy advocates are among the many who share Grothaus’s concerns about deepfake abuse. Yet none are interviewed or even represented in any real depth. Unbelievably, nor are women whose likenesses have been stolen and edited into pornography. Considering that this has dominated public discussion of deepfaking, and that there are multiple charitable foundations in the US alone which offer victim support and resources – Deeptrust Alliance and EndTAB, to name two – this is a baffling oversight. Although Grothaus clearly understands what a hateful offence it is to steal a woman’s persona and force it into hardcore pornography, his decision to describe one such video in unnecessary detail, including the named identity of its celebrity victim, is an unpleasant misjudgment.
“I like to tell people’s stories,” Grothaus writes, and there is certainly one person whose stories he finds particularly mesmerising. Grothaus writes about himself on what feels like every other page, often at length and regardless of whether his personal experiences tell us anything about deepfakes (they usually don’t). At its most egregious, this takes the form of him repeatedly name-dropping his novel, or directly quoting people praising the book to him.
Buried beneath the padding and the stories about himself, however, is a serious and legitimate warning. Technology’s inexorable advance means that deepfakes are vastly easier to create today than they were just two years ago. In two years’ time, it will be easier still. Serious discussion must be had about how abuse of the technology can be curtailed, and Grothaus notes early discussions among US legislators as to where the line should be drawn. However, by relying on hypothetical political disasters or deepfakes he has personally commissioned, Grothaus makes an unintended interesting point. Contrary to widespread anxiety about the technology, as of 2022 at least, deepfakes are not yet wreaking mass havoc, despite the requisite software being available online and the cost of the necessary computer hardware not dissimilar to that of a reasonable gaming PC.
Quite why this might be is an open question. One possible explanation could be that in cases where a story turns on video evidence (say, footage of the prime minister’s spokesperson joking about a staff party mid-pandemic), such footage is usually corroborated, by witnesses, documents or other evidence. Another, suggested by The Guardian’s technology editor Alex Hern, is that humans have spent thousands of years being primed for the possibility that their fellow men and women might be lying to them. Though AI-accelerated, deepfakes are essentially nothing new.
Grothaus does briefly develop this into a persuasive argument that public familiarity with the concept of deepfakes might enhance the so-called “liar’s dividend”, whereby dishonest individuals can dismiss any damning video as fake, hoping that people will be too cynical to try to assess the truth. He cites real instances of this, and in a time of declining trust in news organisations, our industry would do well to consider how it can convince readers and viewers that cutting-edge digital fakes are not creeping into newsrooms.
