Download Artificial Illusions: Deepfakes as Speech and more Study notes Mathematics in PDF only on Docsity!
Intersect , Vol 14, No 3 (202 1 )
Artificial Illusions: Deepfakes as Speech
Nicolas Graber-Mitchell
Amherst College Deepfakes, a new type of artificial media created by sophisticated machine learning algorithms, present a fundamental epistemological problem to society: How can we know the truth when seeing and hearing are not believing? This paper discusses how deepfakes fit into the category of illusory speech, what they do in society, and how to deal with them. Illusions present an alternate reality, much like a lie, but they also contain evidence for that reality. Some illusions, like games of tag and magic tricks, are harmless and fun. Others, like counterfeit coins and deepfakes, harm others and are deeply convincing. For example, the most common use for deepfake technology is to produce pornographic videos of women who never consented. After strangers attacked them in this way, women reported feeling violated and living in a state of constant “visceral fear.” Pornographic deepfakes — most often deployed against women — abridge their targets’ sexual agency and privacy, contributing to inequality and enabling intimate partner abuse, workplace sexual harassment, and other discrimination and hate. Deepfakes also pose a threat in politics and society more generally. In addition to allowing malicious actors to produce convincing, illusory disinformation, their increased use may lead to a general inability to discern the truth. In the face of the deep and distressing harms that deepfakers cause to women and the danger that they present to democracy, this paper argues for new civil and criminal penalties for deepfakers as well as new regulations and liabilities for internet platforms that host their work.
Introduction
In December 2017, a staff writer at Vice discovered a pornographic video of Gal Gadot, an Israeli actress of Wonder Woman fame, having sex with her supposed stepbrother (Cole, 2017). A user with the screenname ‘deepfakes’ created and posted the video on the link-sharing website Reddit along with other videos of other celebrities having sex. The only problem is that Gal Gadot never participated in a pornographic film. Instead, what the reporter at Vice found was the product of a new type of artificial intelligence that swaps one person’s face with another in a video. Depending on the sophistication of the artificial intelligence one uses, the resulting videos can mimic details as tiny as mouth movements
and facial expressions — and appear disturbingly realistic. The original Reddit user ‘deepfakes’ did not employ teams of CGI artists, huge banks of computers, or even cutting-edge technology to produce his illusory pornography. Instead, he used nothing but a consumer-grade computer, publicly available software, and digital images of his targets. Eventually, the manufactured videos themselves became known as “deepfakes.” As deepfake technology spread around the internet, other people discovered new uses for it, like generating audio that mimics the speech of famous politicians (Gholipour, 2017) and creating illusory videos of President Obama (Choi, 2017). Since deepfakes are illusions, and present an alternate reality where people do things they never did and say things they never said, they pose a unique problem to our society and inflict unique harms on individuals. In pornography, deepfakes are essentially nonconsensual sex, and in politics, deepfakes make it impossible for voters to discern the truth, among several other harms. When presented with these troubling videos, what should we do to minimize and eliminate their harms? Since deepfakes are speech, what does the First Amendment prevent government from requiring of deepfake creators and internet platforms? In the rest of this essay, I will explain what deepfakes are and why they are so effective, how they act on the world, the specific harms they cause, and what we can do about it.
What is a deepfake?
Deepfakes are videos, audio, or images of human beings generated, either in part or totally, by advanced artificial intelligence networks. Unlike other types of edited, manipulated media, like airbrushed or photoshopped pictures, deepfakes do not require huge amounts of careful human effort. While deepfaking is a skill, and one’s deepfakes become more and more convincing with practice, even laypeople can create them. Deepfakes take advantage of deep learning, a field of artificial intelligence that has grown in the last decades. Deep learning algorithms consist of a network of linked “neurons” that each receive many inputs in the form of numbers, process those inputs according to a randomly generated weight function, and then output a single number that other neurons in the network use as input. To “train” the algorithm, users feed the network huge amounts of training data along with expected results. The network processes the data and compares its output with the expected results to tweak the weight functions of its neurons. Eventually, after processing enough training data, the deep learning network — also known as a neural network — can successfully process data similar to, but not in, its training set. In 2014, a team of machine learning experts developed a new approach to machine learning known as the generative adversarial network (GAN) (Pan et al., 2019). To produce better results, GANs pair a generating neural network with a “discriminator,” an entirely separate network designed to detect bad results. Then, the generator and the discriminator
faces and place them into already-recorded surrounding video, it is not difficult to imagine near-future GANs that build entire scenes, including personalized human body movements, out of whole cloth. As more time passes, deepfake technology and its potential for illusion only grows.
Illusion
Illusion is the act of illustrating the world contrary to reality. In this way, deepfakes are illusion. They portray something as occurring in real-life when, in fact, it never occurred. When a magician pulls a coin from behind your ear, they do the same — they seek to convince you, through the absolute evidence of the coin’s existence, that it came from behind your ear. Likewise, optical illusions convince you that you are seeing something that does not exist, such as movement on a static page. Falling for an illusion leaves one misled or deceived, either convinced of an untruth or unsure as to the actuality of what they have perceived. When we misapprehend after seeing an illusion, we do not necessarily fail to understand the real world. Rather, we may perceive and understand an alternate world, one that differs from the real world in at least some ways. Depending on how much we hold onto our preconceptions of a non- illusory world, we may doubt illusion, but the hallmark of a good illusion is its ability to introduce even a tiny amount of disbelief into what we thought was true before. After the magician makes a volunteer disappear, they force their audience to grapple with the possibility that the volunteer actually did disappear through the arcane, mysterious forces of magic rather than a clever trick of attention. In fact, illusion goes hand in hand with the suspension of disbelief. Only someone who wants to be bored enters a magic show with the object of doubting the illusions onstage and never giving themselves over to amazement. Likewise, we do not settle into movie theater seats only to gripe about how the computer-generated graphics of monsters and aliens do not accurately portray reality. We enter the theater with the full knowledge that for the next two hours, our enjoyment hinges on believing the illusions beaming out of the screen. Illusion, writ large, is not nefarious or evil. Instead, illusion is fun.^1 Stories, whether told through illusory graphics or not, play off our ability to fall into a world of untruth and fiction, forgetting our surroundings and losing ourselves within an alternate reality. When we convince ourselves that the floor is lava and jump from couch to table in order to avoid burning in the molten rock, we willingly deceive ourselves — and not pathologically, but for fun! Pretending is self-illusion. When we pretend, we choose to misapprehend the world in order to enjoy an alternate reality. In fact, all games are illusion: in a schoolyard game of tag, what does it mean to be “it” if being “it” is not an illusion? If the participant who is “it” (^1) Understanding illusion as play is borne out by its etymology. Illusion comes from the Latin lūdere , meaning to play, to amuse, to mock, to mimic, to tease, to deceive, and to trick, among others.
no longer wants to be “it,” they can simply cease to play — cease to pretend, cease to weave self-illusion — and that alternate reality disappears. Yet people sometimes use illusions for nefarious purposes. Like the magician who produces a coin from behind your ear, some will produce illusory coins and say they are real to manufacture money where there was none before. These coins look and feel precisely the same as their authentic counterparts, but they have no legal value. Counterfeit documents mimic the signs and seals of real documents in order to produce an illusion of authority. Counterfeit paintings, down to their very brushstrokes and paints, pass themselves off as the real deal though they are mere illusions. Similarly, cleverly manipulated images construct an illusory reality in which events that did not happen in the real world happened. The grammar of illusions highlights their internal mechanism. Illusions speak for themselves. The counterfeit coins do the work of maintaining the illusion, not the counterfeiter. By the time that a shopkeepers asks if certain coins are counterfeit, the illusion has already failed. Likewise, a counterfeit painting mimics authenticity down to its very materials because the word of its seller has little bearing when it comes to maintaining the illusion. Instead, illusions require no outside explanation. They contain their own evidence. Realistic computer imagery convinces us it is real because we are seeing it with our own eyes, and a magician’s tricks would not mislead us in the slightest if we could not perceive them. In other words, illusions do not merely describe a world contrary to reality. They illustrate and perform it. As speech, they actually construct that world just like other performatives, such as saying “I do” in a wedding (Austin, 1975). In the words of countless literature teachers, illusions show us an alternate world instead of telling us of it. It is here that deepfakes snugly fit in to the field of illusion. A deepfake is an illusion, not a lie. It tells a lie, but it is not one itself.^2 If a liar says something that is not true, they necessarily have no evidence for their claim unless they construct an illusion to support what they say. In that case, the liar’s speech is secondary. It is the illusion that does the heavy lifting of popping a new, alternate reality into existence. A deepfake that purports to show President Obama cursing, making pop culture, references, and warning the world about deepfakes differs categorically from a statement that claims that happened. The deepfake is an illusion, whereas the statement is a lie. The deepfake counterfeits Barack Obama’s speech, assuming his guise and taking his voice in order to present a claim. The lie, however, is just that: a lie. It does not convince us of its truth; the video of Obama saying those things does. (^2) Even though an illusion is not a lie itself, it is not an illusion if it does not tell a lie or somehow evidence something that is not real. Hence, illusions are always somehow made in relation to the truth. A self-illusion that the floor is lava is not an illusion if the floor is actually lava — nor is the situation enjoyable.
their very form. Deepfakes are the epistemological equivalent of the nuclear bomb except this time everyone has the launch codes (Johnson & Diakopoulos, 2021).
Deepfakes in the wild
Pornography
As mentioned in the introduction, the first-noted and most widespread use of deepfakes is in pornography (Ajder et al., 2019, pp. 1–2). From Gal Gadot, a renowned celebrity, to ordinary women living non-descript lives, anyone is subject to potential deepfake porn if enough pictures of their face exist online. About a month after Vice’s Samantha Cole discovered deepfakes on Reddit, her internet sleuthing turned up countless videos, made by a variety of users, that inserted celebrities’ faces into pornography (Cole, 2018). Her articles’ titles, both of which play off the double meaning of “fucked,” clarify what is at stake in deepfake pornography.^4 When your face is swapped into a pornographic video, the results are just as if you had participated in that video. Your sexual agency is compromised as you are shown, absolutely realistically, engaging in any number of sex acts. It becomes nearly impossible to prove you did not participate, since video evidence of your participation exists right there. Even if you do manage to prove it, the video has already affected your reputation, your mental health, and your sexual agency. Deepfake pornography does not merely depict its targets sexually. Rather, it manufactures a convincing illusion of sex that is as real in its effects as actual sex. An illusion is real when it succeeds. Counterfeit money becomes real in every meaningful way when it is accepted in place of authentic coins. Illusory, deepfake pornography becomes real pornography when it acts on the world in the same way that real pornography would. According to Danielle Keats Citron, deepfake sex videos “hijack people’s sexual and intimate identities” and abridge their sexual privacy and agency (Citron, 2019, p. 1921). Though she writes that deepfake pornography is not the same as the nonconsensual distribution of explicit images (commonly known as “revenge porn”), she ascribes the same harms as revenge porn, sextortion, “up-skirt” photos, and nonconsensual recording to deepfakes. All of these acts invade their target’s sexual privacy, traumatizing them in the process. Citron’s examples include victims of deepfakes and other invasions of sexual privacy feeling unable to walk outside their homes without fearing that someone will recognize them from a pornographic video, suffering from recurrent feelings of exposure and vulnerability, and living in a state of “visceral fear” (Citron, 2019, pp. 1924–1926). Women whose faces appeared in pornographic videos made by ex-partners, unknown abusers, and random men in search (^4) The first meaning uses “fucked” to mean that one is unable to recover from a horrible thing, such as being “finished” or “done for.” The second is the past participle of the verb “to fuck.”
of a good target suffered severe, lasting emotional harm; psychological distress; reputational harm; and even job loss after those videos became public. Deepfake pornography does not affect all people equally. Though the technology is effective against men and women alike, it is most commonly against women when it comes to pornography (Ajder et al., 2019, p. 2). As Citron points out, other acts that rip through one’s sexual privacy, like nonconsensual pornography and cyber-stalking, are similarly gendered (Chesney & Citron, 2019, p. 1773). As such, sexual deepfakes fall squarely into Catharine MacKinnon’s framework of sex-based group defamation-as-discrimination (MacKinnon, 1993, p. 99). In addition to the individual harms of deepfake pornography, illusory sex videos also constitute a gendered attack against women in general that manifests itself in our social reality. For instance, the general threat that anyone may make deepfake sex videos of anyone else constrains the actions of women who want to avoid those harms. Moreover, the power that men have to alienate the sexual privacy and agency of anyone they meet structures every interaction and permeates every space, including the home (Dodge & Johnstone, 2018). Deepfakes enable and sharpen intimate partner abuse, workplace sexual harassment, and other discrimination and hate. MacKinnon’s ideas also illustrate the nature of deepfaked sex videos themselves. A deepfake video is not a depiction of imagined nonconsensual sex, with imagined harms. It is a violent attack on one’s sexual agency. It is as real as sexual assault and as impactful as it too. Just as rape survivors live through daily re-traumatization simply by existing in their bodies, victims of deepfake pornography feel exposed, vulnerable, and taken advantage of by very act of being seen by others. The existence of an illusion of sex becomes as powerful as sex itself.
Political misinformation and disinformation
In 2018, a few months after Samantha Cole broke the story about deepfake porn, former president Barack Obama warned the world about the possibilities of deepfakes turning our democracy into a “fucked-up dystopia” according to a video of his statement from BuzzFeed News (Sosa, 2018). However, Obama never said those words, at least not in a public address. The end of the video reveals the illusion: Obama shrinks to one half of the screen and Jordan Peele, a well-known director, actor, and Obama impersonator, appears in the other half. The rest of the video subjects you to a surreal experience in which Jordan Peele’s voice is Obama’s and Obama’s is Jordan Peele’s. Both make the mouth movements required to produce the speech, and both make hand gestures that could reasonably accompany the audio. While this deepfake used Obama’s visage solely to communicate an educational point and drum up discussion, and clearly did not constitute misinformation, it captures the most obvious problems that deepfakes pose
claim that real videos are actually fake. Dubbed the liar’s dividend, the same dynamic motivates Trump’s repeated fake news claims. Since many voters distrust the media, Trump is able to escape accountability for his actions by claiming that the reporters who broke the story are lying (Chesney & Citron, 2019, p. 1785). Widespread counterfeit evidence only makes it easier for liars to deny true events. The second object of harm that Green identified is the electoral process more broadly. Deepfakes and malicious political illusions reduce voters’ faith in democracy and self-governance (Green, 2019, p. 1460). If enough people become convinced that it is impossible to find the truth, the suppositions underlying democratic government fall apart. The effects of deepfakes on individual voters, taken to their limit, become the dissolution of democratic society itself. Deepfakes’ inherent politics are anti- democratic. Moreover, deepfakes attack the basic institutions of democracy through other avenues than the ballot box. For example, when a journalist encounters video or audio evidence of an important story, how should they treat it when there is a possibility it is a deepfake? Though news agencies fact-check everything they are given, they also rely heavily on the fact that until now, it has been almost impossible to produce convincing illusory audio and video. Even the possibility of a news agency succumbing to a deepfake and spreading its illusion might chill its enthusiasm to fulfill its fact-finding mission (Chesney & Citron, 2019, p. 1784). Repressive regimes have long tried to use illusory images to discredit political opponents, though these illusions can backfire in an interconnected world (Farid, 2011; Venger, 2016). However, deepfakes are a more convincing illusion than single images, and it is already easy for a well-funded government propaganda agency to manufacture humiliating, ruinous deepfakes of individuals who pose political problems. Such video and audio may very well make it extraordinarily difficult for opposition movements to get off the ground. Deepfakes’ other harms to democracy include sowing division, undercutting public safety, constraining diplomacy, and attacking national security (Chesney & Citron, 2019, pp. 1780–1784). In any situation where someone saying something incendiary would cause harm, deepfakes have the potential to manufacture that harm. Examples of this sort of situation abound. In light of the mass protests against police brutality in the summer of 2020, Chesney and Citron’s examples of a deepfaked police chief yelling racial slurs or a deepfaked community leader ordering violence against police officers are especially convincing. Deepfakes also pose a problem for the arbiters of truth in our democracy: the courts. Besides their ability to convince individual judges of different realities, a problem that scholar Richard Hasen dubs “siloed justices,” deepfakes present an epistemological and practical problem during lawsuits (Hasen, 2019, pp. 563–566). If any video at all may be faked, how is it possible for judges and juries to discern true evidence
from falsified evidence? Though complex and invasive solutions that involve blockchains and mathematical verification may work in some cases, such as police bodycams, some evidence will always be unverifiable (“Decoding Deepfakes,” 2020, p. 22). It remains to be seen if rules of evidence will incorporate deepfake-specific provisions for verifying the authenticity of video and audio evidence, but such provisions cannot exist if we cannot detect deepfakes. Finally, Green considers deepfakes’ harms to candidates themselves. She argues that counterfeit campaign speech attacks candidates’ dignity and hijacks their identities in much the same way that deepfake pornography acts on its victims. This is not to say that that they suffer invasions of their sexual privacy, though this is indeed possible. A pornographic deepfake of a woman who is running for office would most likely cost her votes, harming her political campaign at the same time as it abridges her sexual privacy and agency. In general, however, I claim that some of the harm of illusory speech involves the loss of identity itself, which also occurs in pornographic illusion. When a malicious actor produces a deepfake of a candidate, they subjugate the candidate’s identity for use in their illusion.^6 They deny that candidate control over their own speech. Speaking with the candidate’s stolen voice, deepfakers can spread false claims that tank their political careers and cause irreparable harm to their reputation. Though the law of defamation typically covers reputational harms from the point of view of the target — and these harms are real — voters also have an interest in using candidates’ reputations to guide their actions, which becomes impossible when deepfakes construct illusory reputations in place of real ones (Heymann, 2011, p. 1376). Deepfakes’ harms to democracy, whether to voters, society, or candidates, only multiply when we consider the accessibility of deepfakes and modern speech. Deepfake technology itself is widespread and uncomplicated, and the social shortcuts provided by social media allow traditional mis- and disinformation to go viral. Those looking to affect politics or even make quick cash find that their fake news spreads like wildfire on social media (Hasen, 2018, pp. 206–208). Since social media platforms perform very little moderation, and these networks are open to everyone, including anonymous, fake accounts, speech becomes “cheap,” a descriptor first coined in a law review article by Eugene Volokh (1995). More than 20 years later, Richard Hasen revisited Volokh’s predictions and found a far darker world — our world — in which the pathologies of cheap speech were partway through tearing down American democracy. As deepfakes grow in popularity and use, they will act like nitrous oxide in the engine of viral fake news, intensifying its effects and speeding its dissemination. (^6) The theft involved is both physical and theoretical. Deepfakers literally steal their targets’ identities by using their images to train the GANs that generate deepfakes.
development and subsidize detection.^8 At the same time, we must not downplay deepfakes’ capacity for harm. Deepfakes already cause actual damage to their targets and society. These harms cry out for regulation. First, I propose, as others have before, that we strengthen and clarify civil remedies to defamation, privacy invasion, and other abuse rooted in deepfakes. Since deepfakes necessarily present falsehoods and often damage their target’s reputation, they are governed by state defamation statutes (“Defamation,” n.d.). Though these laws differ from state to state, NYT v. Sullivan constrains their application on First Amendment grounds, which makes it more difficult for public figures to successfully pursue these defamation cases. However, states should allow private individuals to sue for civil damages related to deepfake defamation, including pornography, that causes harm. Furthermore, states and the Supreme Court should allow public figures to sue for deepfake defamation without having to prove the “actual malice” standard.^9 Illusions that present false statements — especially sophisticated, hard to detect illusions — differ categorically from those false statements themselves (Blitz, 2018, p. 110). It is coherent to say that defamation of a public figure requires proving actual malice on the part of the perpetrator except when they manufacture evidence to support their claim; the counterfeiting involved turns the speech into fraud rather than simply falsehood (Green, 2019, p. 1483). Since satirical and educational deepfakes do not harm their audience through falsehood, they do not qualify as defamation and are not covered by these changes. Of course, it is not always easy to tell if something is satirical. Rather than attempting to understand the deepfake’s intent, courts should award damages to deepfake targets if someone spreads harmful deepfakes without portraying them as satire. A deepfake created by a satirist, when stripped of its context and watermark, could very easily cause harm. However, civil remedies cannot protect the most vulnerable from the harms of deepfakes. As Danielle Citron notes, civil cases require time, money, and most worryingly, identification. Hence, “many victims [of sexual privacy abuses] decline to bring civil suits because they do not want to expose themselves to their attackers any further” (Citron, 2019, p. 1930). Criminal sanctions, while complete with their own set of problems, can provide victims of deepfakes with protection even when they wish to remain anonymous. Particularly when it comes to nonconsensual deepfake pornography, we need criminal penalties to protect victims and “signal the significant harm that such invasions inflict” (Citron, 2019, p. 1931). If states and the federal government do not acknowledge the need for a (^8) Even though reality constrains us, the question of whether we should snap our fingers and eliminate deepfake technology if it were possible is an interesting one. Given the catastrophic societal pathologies that deepfakes may exacerbate in the next couple of years, I lean towards snapping. (^9) As a term of art, actual malice means “knowledge or reckless disregard for the possibility that [something was] false.” (Chesney & Citron, 2019, pp. 1793–1794)
comprehensive sexual privacy statute, as Citron promotes, they should at least ban the creation and distribution of deepfake pornography (as well as other nonconsensual pornography). Catharine MacKinnon’s observation that for such media, what it says is “exactly the measure of [its] harm” supports such a ban (MacKinnon, 1993, p. 22); there is no protection in the First Amendment for invading another’s sexual privacy and using their image as sex, regardless of the fact that such an act does speak. Criminal sanctions for deepfakes of candidates or government officials are equally prudent. Like civil penalties, we should exclude satirical, educational, and other non-damaging deepfakes from these statutes to avoid as many First Amendment issues as possible. Such statutes should focus on the harm done to American democracy and voters in their construction. In particular, a deepfake of a sitting government official has extraordinary potential for harm; just as law proscribes the impersonation of a government officer, deepfakes of them should be illegal as well (Citron, 2019, p. 1937). However, the nature of social media hinders enforcing civil and criminal sanctions against the creators and distributors of deepfakes. Since so many accounts are anonymous, finding an actual perpetrator who can stand trial is not an easy task. Moreover, these websites are international, and a deepfake spread by a bot account somewhere outside the United States damages our democracy and our privacy just as much as a homegrown deepfake. In this gap, we must require internet platforms to act against deepfakes. According to Richard Hasen, “the government likely has the power under the Constitution to mandate a truth-in-labeling law requiring social media platforms [...] to label synthetic media” (2019, p. 549). The federal government should enact regulations requiring social media platforms to label viral deepfakes on their platforms. Though the automatic deepfake detection is not yet possible, human labelers can act as a stopgap measure until we develop better forensic tools. Beyond labeling, Congress should require platforms to remove deepfakes that fit into the civil and criminal sanctions that I discussed above. Though Hasen does not go this far, such a requirement is necessary. Just because we cannot identify a perpetrator in the United States to prosecute for a given pornographic or political deepfake does not mean that the deepfake itself should freely spread around the internet. In particular, we should adopt Bobby Chesney and Danielle Citron’s limiting amendment to Section 230, the federal law that governs internet platforms (2019, p. 1799). In its current form, the law has “evolved into a super- immunity” to legal liability for content that allows internet platforms to “ignore the propagation of damaging deep fakes ”(Chesney & Citron, 2019, p. 1798). By introducing civil liabilities for platforms that continue to spread harmful deepfake videos, the federal government can encourage those platforms to take down illusory, dangerous content. In doing so, we can “minimize the most-serious harms that might follow from user-posted or user-distributed deep fakes” (Chesney & Citron, 2019, p. 1799). To
sophisticated illusory evidence, complicate the diplomatic process, and deepen partisan divides. Mis- and disinformation already poison our democracy without the help of advanced, hard-to-detect illusions. Deepfakes will only exacerbate these problems. To stay on top of the problems posed by deepfakes, we should pass a battery of new laws targeted at their specific, actionable harms. From civil and criminal liability for harmful deepfakes to internet platform regulations, we have First Amendment-compliant legal remedies that will help those most hurt by deepfakes. We also need to discourage deepfaking more broadly, working through social norms in the areas where law cannot help us. A comprehensive social response, with both norms and laws, is the only thing that can reduce the present and future harms of deepfakes. To quote Jordan Peele wearing Barack Obama’s skin, we must take action if we are to escape our impending future: a deepfake-ridden, fucked up dystopia (Sosa, 2018).
References
Ajder, H., Patrini, G., Cavalli, F., & Cullen, L. (2019). The State of Deepfakes: Landscape, Threats, and Impact (p. 27). Deeptrace. https://web.archive.org/web/20201004082457/https://regmedia.co.uk/ 019/10/08/deepfake_report.pdf Austin, J. L. (1975). How to do things with words (J. O. Urmson & M. Sbisá, Eds.; Second edition.). Harvard University Press. Blitz, M. J. (2018). Lies, Line Drawing, and Deep Fake News. Oklahoma Law Review , 71 (1), 59–116. Chesney, B., & Citron, D. K. (2019). Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security. California Law Review , 107 (6), 1753–1820. Choi, C. Q. (2017, July 12). AI Creates Fake Obama. IEEE Spectrum: Technology, Engineering, and Science News. https://spectrum.ieee.org/tech-talk/robotics/artificial-intelligence/ai- creates-fake-obama Citron, D. K. (2019). Sexual Privacy. Yale Law Journal , 128 (7), 1870–
Cole, S. (2017, December 11). AI-Assisted Fake Porn Is Here and We’re All Fucked. Vice. https://www.vice.com/en/article/gydydm/gal-gadot- fake-ai-porn Cole, S. (2018, January 24). We Are Truly Fucked: Everyone Is Making AI-Generated Fake Porn Now. Vice. https://www.vice.com/en_us/article/bjye8a/reddit-fake-porn-app- daisy-ridley Decoding Deepfakes: How Do Lawyers Adapt When Seeing Isn’t Always Believing? (2020, April). Oregon State Bar Bulletin , 80 (6), 7. Defamation. (n.d.). In Wex. Legal Information Institute. Retrieved September 27, 2020, from https://www.law.cornell.edu/wex/defamation Dodge, A., & Johnstone, E. (2018). Using Fake Video Technology To Perpetrate Intimate Partner Abuse (p. 8). Without My Consent. Farid, H. (2011). Photo Tampering Throughout History. https://web.archive.org/web/20110606065418/http://www.cs.dartmout h.edu/farid/research/digitaltampering/ Flynn, D. J., Nyhan, B., & Reifler, J. (2017). The Nature and Origins of Misperceptions: Understanding False and Unsupported Beliefs About Politics. Political Psychology , 38 (S1), 127–150. https://doi.org/10.1111/pops. Frankfurt, H. G. (2005). On bullshit. Princeton University Press. Gholipour, B. (2017, May 2). New AI Tech Can Mimic Any Voice. Scientific American. https://www.scientificamerican.com/article/new- ai-tech-can-mimic-any-voice/ Green, R. (2019). Counterfeit Campaign Speech. Hastings Law Journal , 70 (6), 1445–1490. Hao, K. (2019, June 12). Deepfakes have got Congress panicking. This is
States 2012). https://advance-lexis- com.ezproxy.amherst.edu/api/permalink/cb0d7390-e267- 4725 - 8161 - 8f6eafb3e603/?context= Venger, O. (2016). When Shaming Backfires: The Doublespeak of Digitally-Manipulated Misogynistic Photographs. CyberOrient , 10 (1), 61 – 86. Virginia bans “deepfakes” and “deepnudes” pornography. (2019, July 2). BBC News. https://www.bbc.com/news/technology- 48839758 Volokh, E. (1995). Cheap Speech and What It Will Do. The Yale Law Journal , 104 (7), 1805–1850. JSTOR. https://doi.org/10.2307/ Winner, L. (1980). Do Artifacts Have Politics? Daedalus , 109 (1), 121–
