You may have heard that in recent days a series of deepfake videos appeared on TikTok of a fake Tom Cruise looking very Tom-Cruise-ish all while doing mostly non-Tom-Cruise-ish things. After that series of short videos came out, the parties responsible for producing them, Chris Ume and Cruise impersonator Miles Fisher, put out a compilation video sort of showing how this was all done.As you can see, this was all done in the spirit of educating the public on what is possible with this kind of technology and, you know, fun. Unfortunately, some folks out there aren't finding any fun in this at all. Instead, there is a certain amount of understandable fear for how this technology might disrupt our lives that is leading to less understandable conclusions about what we should do about it.For instance, some folks apparently think that deepfake outputs should be considered the intellectual property of those who are the subjects of the deepfakes.
A recent deepfake of Hollywood star "Tom Cruise" sparked a bombshell after looking very close to real. Now it has been claimed they are on their way to becoming so good, that families of the dead should own the copyright of their loved ones in deepfakes.Lilian Edwards, a professor of law and expert in the technology, says the law hasn't been fully established yet. She believes many will claim they should own the rights, while some may not.She told BBC: "For example, if a dead person is used, such as (the actor) Steve McQueen or (the rapper) Tupac, there is an ongoing debate about whether their family should own the rights (and make an income from it)."
Now, I want to be somewhat generous here, but this is still a terrible idea. Let's just break this down practically. In the interest of being fair, it is understandable that people would be creeped out by deepfake creations of either their dead relatives or themselves. Let's call that a given. But why is the response to that to try to inject some kind of strange intellectual property right into all of this? Why should Steve McQueen's descendants have some right to control this kind of output? And why are we using McQueen and Tupac as the examples here, given that both are public figures? What problem does this solve?The answer would be, I think: control over the likeness rights of a person. But such control is both fraught with potential for overreach and over-protection coupled with a history of a total lack of nuance in what should not be considered infringing behavior or what is fair use. Techdirt's pages are littered with examples of this. Add to all of this that purveyors of deepfakes are quite often internationally located, anonymous, and unlikely to pay the slightest attention to the kind of image likeness rights being bandied about, and you really have to wonder why we're even entertaining this subject.And then there are the people who think this Tom Cruise deepfake means that soon we'll simply have no functional legal system at all.
The CEO of Amber, a video verification site, believes deepfake evidence will raise reasonable doubt. Mr Allibhai told us: “Deepfakes are getting really good, really fast.“I am worried about both aural/visual evidence being manipulated and also just the fact that when believable fake videos exist, they will delegitimise genuine evidence and defendants will raise reasonable doubt. When the former happens, innocent people will go to jail and when the latter happens, criminals will be set free. Due process will be compromised and a core foundation of democracy is undermined. Judges will drop cases, not necessarily because they believe jurors will be unable to tell the difference: they themselves, and most humans for that matter, will be unable to tell the difference between fact and fiction soon."
Folks, we really need to slow our roll here. Deepfake technology is progressing. And it's not progressing slowly, but nor is it making insane leaps heretofore unforeseen. The collapse of the legal system as a result of nobody being able to tell truth from fiction may well come one day, but it certainly won't be coming as a result of the harbinger of a Tom Cruise deepfake.In fact, you really have to dial in on how the Cruise videos were made to understand how unique they are.
The Tom Cruise fakes, though, show a much more beneficial use of the technology: as another part of the CGI toolkit. Ume says there are so many uses for deepfakes, from dubbing actors in film and TV, to restoring old footage, to animating CGI characters. What he stresses, though, is the incompleteness of the technology operating by itself. Creating the fakes took two months to train the base AI models (using a pair of NVIDIA RTX 8000 GPUs) on footage of Cruise, and days of further processing for each clip. After that, Ume had to go through each video, frame by frame, making small adjustments to sell the overall effect; smoothing a line here and covering up a glitch there. “The most difficult thing is making it look alive,” he says. “You can see it in the eyes when it’s not right.”Ume says a huge amount of credit goes to Fisher; a TV and film actor who captured the exaggerated mannerisms of Cruise, from his manic laugh to his intense delivery. “He’s a really talented actor,” says Ume. “I just do the visual stuff.” Even then, if you look closely, you can still see moments where the illusion fails, as in the clip below where Fisher’s eyes and mouth glitch for a second as he puts the sunglasses on.
This isn't something where we're pushing a couple of buttons and next thing you know you're seeing Tom Cruise committing a homicide. Instead, creating these kinds of deepfakes takes time, hardware, skill, and, in this case, a talented actor who already looked like the subject of the deepfake. It's a good deepfake, don't get me wrong. But it is neither easy to make nor terribly difficult to spot clues for what it is.All of which isn't to say that deepfakes might not someday present problems. I actually have no doubt that they will. But as with every other kind of new technology, you're quite likely to hear a great deal of exaggerated warnings and fears compared with what those challenges will actually be.
Summary: In 2012, the Huffington Post did an expos on eating disorder blogs, mainly on the site Tumblr. It discussed the world of thinspo and thinspiration blogs, that focused on building a community around losing unhealthy amounts of weight. In response, Tumblr announced that it was banning self harm blogs, and classified eating disorder blogs among those no longer allowed.Three years later, a study by Munmun De Choudhury discussed how there was still eating disorder information on Tumblr, but that it was mainly split into two different categories: those who were supportive of eating disorders such as anorexia (referred to as proana) as well as communities built up around recovering from eating disorders. One interesting finding of the report was that the recovery groups often used the same keywords and messaging, in an attempt to permeate among the proana groups, in order to try to encourage those with eating disorders to seek support, therapy, and help towards recovery.That same year, Amanda Hess argued in Slate that the rush to ban content about eating disorders on social media (or, in the case of France, where such things were outlawed) was the wrong approach.
"But while we know anorexia can kill, we're not quite sure what happens to people who read about it online. In an article published last month, Canadian criminologists Debra Langan and Nicole Schott could find 'no scholarly evidence' that pro-ana blogs pose a threat to their audiences. If they do, there's no proof that any of our social remediescensorship, PSAs, or prison timedo anything to help. These campaigns are most obviously effective at flattering the egos of the lawmakers and tech execs who champion them. When a girl searches Tumblr for a pro-ana-adjacent term like #thinspo or #thighgap now, Tumblr intercepts her request with bland concern ('Everything OK?'), then advises her to check out the cutesy motivational messaging on the National Eating Disorders Association's Tumblr instead. However the girl responds, Tumblr can feel satisfied it's performed its civic responsibility. The strategy recalls the one favored by a 19th-century doctor who believed that reading novels caused hysteria in women: He counseled men to confiscate their wives' fiction and replace it with a book on 'some practical subject,' like 'beekeeping.'"
The following year, De Choudhury and other authors released another study detailing how pro-eating disorder groups would get around social media blocks on their content by changing words around, or slightly misspelling them, suggesting that the out-and-out blocking method was likely to continue to be ineffective.Another article suggested that the blocks almost made it easier to find information about eating disorders, because dozens of new hashtags were created for the community, rather than just a few before social media sites began to ban such content.A study released in the peer-reviewed New Media & Society journal in 2018 highlighted how easy it appeared to be for users to get around attempts to block content regarding certain eating disorders. The researcher, Ysabel Gerrard, looked mainly at Pinterest, Tumblr, and Instagram, finding that while all three had some policies in place regarding eating disorder information, it was not difficult to find groups or sites dedicated to such information.As summarized by Wired:
"She immediately found that Instagram's pro-ED hashtag ban has an easy workaround: You can search for people who have the keywords in their usernames, just not hashtagged in their posts. She identified 74 public accounts that had terms like 'proana,' 'proanorexia,' or 'thighgap' in their names or bios and who also posted pro-ED content. Then, she analyzed 1,612 of their postsonly 561 of which had hashtagsby cataloguing the content of the image and its caption."On Tumblr, Gerrard followed a number of terms related to pro-ED content, like 'thinspo,' 'proana,' and 'bulimic.' Tumblr allows you to follow topics without needing to follow specific users. For example, you can simply follow 'movies' without following any specific user who posts about that topic. Through this method, she found 50 pro-ED blogs and analyzed 20 posts from each, or 1,000 posts total. Only 218 of the posts were tagged."
The report also found that the recommendation algorithms often drove users towards more such content. By saving a few 'proana' blogs, Gerrard found that Tumblr began recommending more. While it did also recommend some recovery blogs, Gerrard found them easy to exclude.
"Once I had followed ED-related terms - anorexia, anorexic, bulimia, bulimic, thinspiration, thinspo, proana, purge, purging - the platform delivered this content to me through my dashboard and also via email. Tumblr showed me relevant posts and suggested a list of users whose accounts I should follow. As some of these terms are not straightforwardly pro-ED (unlike, for example, proana), I was presented with blogs identifying as 'pro-recovery' in their biographies. But I excluded these blogs from the dataset as they were not the focus of my analysis. Tumblr recommended blogs that were, for example, 'big in proana' or 'like' other popular blogs. I identified fifty pro-ED users through this method."
Another study from 2014 argued that even the proana content represented a double-edged sword and might help some of those either with eating disorders, or those at risk, to recognize that what they were exploring was unhealthy.Decisions to be made by Tumblr/Instagram/Pinterest:
How do you deal with information about eating disorders? Is it actually possible to ban it?
How do you distinguish between proana and recovery content?
Are there other interventions that can be done, such as putting up warning labels, or directing users towards recovery resources when they search on certain terms?
How should recommendation algorithms handle information about eating disorders? Does it need to be adjusted to avoid sending people towards content that glorifies such disorders?
Questions and policy implications to consider:
Does banning content of those promoting unhealthy eating disorders actually help prevent eating disorders?
Does reading about eating disorders function as a how-to guide for the vulnerable, or does it help those at risk recognize that risk? If both, how do you balance these two competing forces?
Can pointing people towards recovery content or other helpful resources lead to better outcomes?
Resolution: Various websites struggle with how to deal with eating disorder information and communities. Attempts to ban it have continued to fail, as the various communities continue to figure out ways to route around any ban. The research on the impact of this content remains mixed, however, and there have been some concerns that efforts to ban such content on certain platforms only makes it move to others that are less well organized to handle the issue.The latest is a report that teens who were engaged in eating disorder discussions on Tumblr have now moved to TikTok. However, that same article also notes that, unlike Tumblr, TikTok seems to have a number of users who celebrate healthy eating and living, and that TikTok's algorithm may be inserting such videos mixed in with those discussing more unhealthy eating behavior.Originally published on the Trust & Safety Foundation website.