The recent emergence of machine-manipulated media raises an important societal question: How can we know whether a video that we watch is real or fake? In two online studies with 15,016 participants, we present authentic videos and deepfakes and ask participants to identify which is which. We compare the performance of ordinary human observers with the leading computer vision deepfake detection model and find them similarly accurate, while making different kinds of mistakes. Together, participants with access to the model’s (...) prediction are more accurate than either alone, but inaccurate model predictions often decrease participants’ accuracy. To probe the relative strengths and weaknesses of humans and machines as detectors of deepfakes, we examine human and machine performance across video-level features, and we evaluate the impact of preregistered randomized interventions on deepfake detection. We find that manipulations designed to disrupt visual processing of faces hinder human participants’ performance while mostly not affecting the model’s performance, suggesting a role for specialized cognitive capacities in explaining human deepfake detection performance. -/- . (shrink)

Deepfake technology uses machine learning to fabricate video and audio recordings that represent people doing and saying things they've never done. In coming years, malicious actors will likely use this technology in attempts to manipulate public discourse. This paper prepares for that danger by explicating the unappreciated way in which recordings have so far provided an epistemic backstop to our testimonial practices. Our reasonable trust in the testimony of others depends, to a surprising extent, on the regulative effects of the (...) ever-present possibility of recordings of the events they testify about. As deepfakes erode the epistemic value of recordings, we may then face an even more consequential challenge to the reliability of our testimonial practices themselves. (shrink)

Recent develops in AI technology have led to increasingly sophisticated forms of video manipulation. One such form has been the advent of deepfakes. Deepfakes are AI-generated videos that typically depict people doing and saying things they never did. In this paper, I demonstrate that there is a close structural relationship between deepfakes and more traditional fake barn cases in epistemology. Specifically, I argue that deepfakes generate an analogous degree of epistemic risk to that which is found (...) in traditional cases. Given that barn cases have posed a long-standing challenge for virtue-theoretic accounts of knowledge, I consider whether a similar challenge extends to deepfakes. In doing so, I consider how Duncan Pritchard’s recent anti-risk virtue epistemology meets the challenge. While Pritchard’s account avoids problems in traditional barn cases, I claim that it leads to local scepticism about knowledge from online videos in the case of deepfakes. I end by considering how two alternative virtue-theoretic approaches might vindicate our epistemic dependence on videos in an increasingly digital world. (shrink)

[Author note: There is a video explainer of this paper on youtube at the new work in philosophy channel (search for surname+deepfakes).] -/- It is widely thought that deepfake videos are a significant and unprecedented threat to our epistemic practices. In some writing about deepfakes, manipulated videos appear as the harbingers of an unprecedented _epistemic apocalypse_. In this paper I want to take a critical look at some of the more catastrophic predictions about deepfake videos. I will argue (...) for three claims: (1) that once we recognise the role of social norms in the epistemology of recordings, deepfakes are much less concerning, (2) that the history of photographic manipulation reveals some important precedents, correcting claims about the novelty of deepfakes, and (3) that proposed solutions to deepfakes have been overly focused on technological interventions. My overall goal is not so much to argue that deepfakes are not a problem, but to argue that behind concerns around deepfakes lie a more general class of social problems about the organisation of our epistemic practices. (shrink)

This paper takes up the question of how videographic public announcements (VPAs)---i.e. videos that a wide swath of the public sees and knows that everyone else can see too--- have functioned to mobilize people politically, and how the presence of deepfakes in our information environment stands to change the dynamics of this mobilization. Existing work by Regina Rini, Don Fallis and others has focused on the ways that deepfakes might interrupt our acquisition of first-order knowledge through videos. But (...) I point out that even where every audience member takes a video to be veridical, where first-order knowledge acquisition is secure, an audience aware of deepfakes in their environment will not acquire higher order knowledge in the way that has erstwhile been characteristic for audiences of VPAs. Engaging with ideas from the literatures on public announcement logic, common knowledge, and convention, I enumerate a variety of ways in which we should expect this absence of higher order knowledge to throw up barriers to political mobilization. I go on to apply my analysis of VPAs to the mechanisms by which an uptick in publicly available videos of police brutality over the last decade, mediated by camera phones and social media, was responsible for the largest mass protests in US history in summer 2020. This makes vivid the stakes of the transformation in our mobilizing environment that I've claimed deepfakes effect: where we lose the common-knowledge-generating effects of VPAs, there are fresh obstacles to this sort of mass mobilization. (shrink)

In recent years, a number of philosophers have turned their attention to developments in Artificial Intelligence, and in particular to deepfakes. A deepfake is a portmanteau of ‘deep learning' and ‘fake', and for the most part they are videos which depict people doing and saying things they never did. As a result, much of the emerging literature on deepfakes has turned on questions of trust, harms, and information-sharing. In this paper, I add to the emerging concerns around (...) class='Hi'>deepfakes by drawing on resources from vice epistemology. As deepfakes become more sophisticated, I claim, they will develop to be a source of online epistemic corruption. More specifically, they will encourage consumers of digital online media to cultivate and manifest various epistemic vices. My immediate focus in this paper is on their propensity to encourage the development of what I call ‘intellectual cynicism'. After sketching a rough account of this epistemic vice, I go on to suggest that we can partially offset such cynicism – and fears around deceptive online media more generally – by encouraging the development what I term a trained ‘digital sensibility'. This, I contend, involves a calibrated sensitivity to the epistemic merits of online content. (shrink)

Deepfakes are algorithmically modified video and audio recordings that project one person’s appearance on to that of another, creating an apparent recording of an event that never took place. Many scholars and journalists have begun attending to the political risks of deepfake deception. Here we investigate other ways in which deepfakes have the potential to cause deeper harms than have been appreciated. First, we consider a form of objectification that occurs in deepfaked ‘frankenporn’ that digitally fuses the parts (...) of different women to create pliable characters incapable of giving consent to their depiction. Next, we develop the idea of ‘illocutionary wronging’, in which an individual is forced to engage in speech acts they would prefer to avoid in order to deny or correct the misleading evidence of a publicized deepfake. Finally, we consider the risk that deepfakes may facilitate campaigns of ‘panoptic gaslighting’, where many systematically altered recordings of a single person's life undermine their memory, eroding their sense of self and ability to engage with others. Taken together, these harms illustrate the roles that social epistemology and technological vulnerabilities play in human ethical life. (shrink)

Deepfake technology can be used to produce videos of real individuals, saying and doing things that they never in fact said or did, that appear highly authentic. Having accepted the premise that Deepfake content can constitute a legitimate form of expression, it is not immediately clear where the rights of content producers and distributors end, and where the rights of individuals whose likenesses are used in this content begin. This paper explores the question of whether it can be plausibly argued (...) that Deepfake content involving the likenesses of real individuals violates the rights of these individuals. (shrink)

Would suppressing deepfakes violate freedom of expression norms? The question is pressing because the deepfake phenomenon in its more poisonous manifestations appears to call for a response, and automated targeting of some kind looks to be the most practically viable. Two simple answers are rejected: that deepfakes do not deserve protection under freedom of expression legislation because they are fake by definition; and that deepfakes can be targeted if but only if they are misleadingly presented as authentic. (...) To make progress, following a discussion of why freedom of expression deserves protection in a democracy, the question is reframed. At issue is not whether the arrival of deepfakes brings new and potentially serious dangers (it does), nor whether these dangers call for measures that potentially limit freedom of expression (they do), but whether the need for such measures raises any new and unfamiliar freedom-of-expression challenges. The answer to that question, surprisingly, is no. The balancing act needed to cope with the arrival of deepfakes brings plenty of difficulties, certainly, but none of the measures likely to be effective in tackling deepfake harms raises freedom-of-expression concerns that aren’t familiar from consideration of non-deepfake harms. In that respect, at least, the arrival of deepfakes makes no difference. (shrink)

AI, especially in the case of Deepfakes, has the capacity to undermine our confidence in the original, genuine, authentic nature of what we see and hear. And yet digital technologies, in the form of databases and other detection tools also make it easier to spot forgeries and to establish the authenticity of a work. Using the notion of ectypes, this paper discusses current conceptions of authenticity and reproduction and examines how, in the future, these might be adapted for use (...) in the digital sphere. (shrink)

In this paper, I draw a distinction between two types of deepfake, and unpack the deceptive strategies that are made possible by the second. The first category, which has been the focus of existing literature on the topic, consists of those deepfakes that act as a fabricated record of events, talk, and action, where any utterances included in the footage are not addressed to the audience of the deepfake. For instance, a fake video of two politicians conversing with one (...) another. The second category consists of those deepfakes that direct an illocutionary speech act - such as a request, injunction, invitation, or promise - to an addressee who is located outside of the recording. For instance, fake footage of a company director instructing their employee to make a payment, or of a military official urging the populace to flee for safety. Whereas the former category may deceive an audience by giving rise to false beliefs, the latter can more directly manipulate an agent's actions: the speech act's addressee may be moved to accept an invitation or a summons, follow a command, or heed a warning, and in doing so further a deceiver's unethical ends. (shrink)

Everyday (mis)uses of deepfakes define prevailing conceptualizations of what they are and the moral stakes in their deployment. But one complication in understanding deepfakes is that they are not photographic yet nonetheless manipulate lens-based recordings with the intent of mimicking photographs. The harmfulness of deepfakes, moreover, significantly depends on their potential to be mistaken for photographs and on the belief that photographs capture actual events, a tenet known as the transparency thesis, which scholars have somewhat ironically attacked (...) by citing digital imaging techniques as counterexamples. Combining these positions, this paper sets out two core points: (1) that conceptions about the nature of photography introduce imperatives about its uses; and (2) that popular cultural understandings of photography imply normative ideas that infuse our encounters with deepfakes. Within this, I further raise the question of what moral ground deepfakes occupy that allows them to have such a potentially devastating effect. I show that answering this question involves reinstating the notion that photographs are popularly conceived of as transparent. The rejoinder to this argument, however, is that to take the sting out of deepfakes we must, once again, become skeptical of the veracity of all images, including photoreal ones. This kind of critical mindedness was warranted before the invention of photography for both pictorial imprints and written accounts from various media sources. Given this, along with the fact that photographic trickery is nothing new, deepfakes need not push us into a post-truth epistemic abyss, for they imply a decidedly old turn. (shrink)

Deepfakes pose a multi-faceted threat to the acquisition of knowledge. It is widely hoped that technological solutions—in the form of artificially intelligent systems for detecting deepfakes—will help to address this threat. I argue that the prospects for purely technological solutions to the problem of deepfakes are dim. Especially given the evolving nature of the threat, technological solutions cannot be expected to prevent deception at the hands of deepfakes, or to preserve the authority of video footage. Moreover, (...) the success of such technologies depends on institutional trust that is in short supply. Finally, outsourcing the discrimination between the real and the fake to automated, largely opaque systems runs the risk of undermining epistemic autonomy. (shrink)

Deepfakes are audio, video, or still-image digital artifacts created by the use of artificial intelligence technology, as opposed to traditional means of recording. Because deepfakes can look and sound much like genuine digital recordings, they have entered the popular imagination as sources of serious epistemic problems for us, as we attempt to navigate the increasingly treacherous digital information environment of the internet. In this paper, I attempt to clarify what epistemic problems deepfakes pose and why they pose (...) these problems, by drawing parallels between recordings and our own senses as sources of evidence. I show that deepfakes threaten to undermine the status of digital recordings as evidence. The existence of deepfakes thus encourages a kind of skepticism about digital recordings that bears important similarities to classic philosophical skepticism concerning the senses. However, the skepticism concerning digital recordings that deepfakes motivate is also importantly different from classical skepticism concerning the senses, and I argue that these differences illuminate some possible strategies for solving the epistemic problems posed by deepfakes. (shrink)

The interpretation of many texts in the everyday world is concerned with their truth value in relation to the reality around us. The recent publication experiments with computer-generated texts have shown that the distinction between true and false, or reality and fiction, is not always clear from the text itself. Essentially, in today’s media space, one may encounter texts, videos or images that deceive the reader by displaying nonsensical content or nonexistent events, while nevertheless appearing as genuine human-produced messages. This (...) article outlines certain problems with artificial intelligence (AI)-generated content, and frames the issue as a problem of recognising its proper referential reality. Examples include generative texts, deepfakes and their functioning in contemporary culture. The article makes use of the concepts of mimicry and nonsense to reveal the elements and counterparts in the communicative processes involving generated media. (shrink)

Deep learning algorithms are rapidly changing the way in which audiovisual media can be produced. Synthetic audiovisual media generated with deep learning—often subsumed colloquially under the label “deepfakes”—have a number of impressive characteristics; they are increasingly trivial to produce, and can be indistinguishable from real sounds and images recorded with a sensor. Much attention has been dedicated to ethical concerns raised by this technological development. Here, I focus instead on a set of issues related to the notion of synthetic (...) audiovisual media, its place within a broader taxonomy of audiovisual media, and how deep learning techniques differ from more traditional approaches to media synthesis. After reviewing important etiological features of deep learning pipelines for media manipulation and generation, I argue that “deepfakes” and related synthetic media produced with such pipelines do not merely offer incremental improvements over previous methods, but challenge traditional taxonomical distinctions, and pave the way for genuinely novel kinds of audiovisual media. (shrink)

This article is devoted to the topical aspects of the transformation of society, science, and man in the context of E. László’s work «Macroshift». The author offers his own attempt to consider the attributes of macroshift and then use these attributes to operationalize further analysis, highlighting three essential elements: the world has come to a situation of technological indistinguishability between the natural and the artificial, to machines that know everything about humans. Antiquity aspired to beauty and saw beauty in realistic (...) art, but technology made indistinguishability possible, bringing the risks of deepfakes and post-truth. Following J. Bernal’s logic, the author distinguishes between different types of the future, comparing them with the Kantian triad of ethical questions: (1) What can we know? (2) What should we do? (3) What can I hope for? Under these conditions, the fact that technology has become a real tool of geopolitics is no longer just a matter for discussion, but a subject of deep philosophical consideration. AI is becoming not only an opportunity, but also a threat, on a scale that has never been seen before in our civilizational history. Therefore, the former, linear development of science gives way to nonlinearity. Society in the macroshift will be radically transformed, and the author identifies five main transformations of the current macroshift. The post-nonclassical approach in the analysis of scenarios for modeling the future, both accidentally possible and preferred, is becoming the most important tool of futurology and prognostics. The author describes the future around five social trends that will determine the economy and society of the coming decades. In addition, opportunities for the future development of science are proposed, subject to significant penetration of AI technologies. These technologies will have a fundamental impact on all aspects of our society, economy, and human existence. (shrink)

In the story of Moses and Pharaoh, the magicians who were there became the first believers in Moses because they believed in the miraculous power of Moses, which was from Allah. In fact, those sticks (sticks of magicians) did not turn into snakes, but were seen by others as snakes. When Moses dropped his stick and turned into a snake, the sorcerers realized that the stick had become a real snake, and that is why they believed Moses. Today, this magic (...) is known as “Project Blue Beam” and “Deepfake” etc. These are real examples of the same sticks that were thrown by magicians. These technologies are not real, and you see them with your eyes the way you think they are real. In this paper, we will examine the miracles of Imam Mahdi. (shrink)

Generative artificial intelligence has given us synthetic media that are increasingly easy to create and increasingly hard to distinguish from photographs and videos. Whereas an existing literature has been concerned with how these new media might make a difference for would-be knowers—the viewers of photographs and videos—I advance a thesis about how they will make a difference for would-be communicators—those who embed photos and videos in their speech acts. I claim that the presence of these media in our information environment (...) reduces our ability to show one another things, even as it may increase our resources for telling. And I argue that this has consequences beyond the disruption of knowledge acquisition; showing is a way that we preserve relational equality through superficial asymmetries in political communication, and thereby express respect for our audiences. If synthetic media reduce our options for showing, they then interfere in the way that we manage our relationships in the context of collective political action. (shrink)

This book argues that misinformation poses a multi-faceted threat to knowledge, while arguing that some forms of content moderation risk exacerbating these threats. It proposes alternative forms of content moderation that aim to address this complexity while enhancing human epistemic agency. The proliferation of fake news, false conspiracy theories, and other forms of misinformation on the internet and especially social media is widely recognized as a threat to individual knowledge and, consequently, to collective deliberation and democracy itself. This book argues (...) that misinformation presents a three-pronged threat to knowledge. While researchers often focus on the role of misinformation in causing false beliefs, this deceptive potential of misinformation exists alongside the potential to suppress trust and to distort the perception of evidence. Recognizing the multi-faceted nature of this threat is essential to the development of effective measures to mitigate the harms associated with misinformation. The book weaves together work in analytic epistemology with emerging empirical work in other disciplines to offer novel insights into the threats posed by misinformation. Additionally, it breaks new ground by systematically assessing different forms of content moderation from the perspective of epistemology. Misinformation, Content Moderation, and Epistemology will appeal to philosophers working in applied and social epistemology, as well as scholars and advanced students in disciplines such as communication studies, political science, and social psychology who are researching misinformation. (shrink)

What separates the unique nature of human consciousness and that of an entity that can only perceive the world via strict logic-based structures? Rather than assume that there is some potential way in which logic-only existence is non-feasible, our species would be better served by assuming that such sentient existence is feasible. Under this assumption, artificial intelligence systems (AIS), which are creations that run solely upon logic to process data, even with self-learning architectures, should therefore not face the opposition they (...) have to gaining some legal duties and protections insofar as they are sophisticated enough to display consciousness akin to humans. Should our species enable AIS to gain a digital body to inhabit (if we have not already done so), it is more pressing than ever that solid arguments be made as to how humanity can accept AIS as being cognizant of the same degree as we ourselves claim to be. By accepting the notion that AIS can and will be able to fool our senses into believing in their claim to possessing a will or ego, we may yet have a chance to address them as equals before some unforgivable travesty occurs betwixt ourselves and these super-computing beings. (shrink)

What separates the unique nature of human consciousness and that of an entity that can only perceive the world via strict logic-based structures? Rather than assume that there is some potential way in which logic-only existence is non-feasible, our species would be better served by assuming that such sentient existence is feasible. Under this assumption, artificial intelligence systems (AIS), which are creations that run solely upon logic to process data, even with self-learning architectures, should therefore not face the opposition they (...) have to gaining some legal duties and protections insofar as they are sophisticated enough to display consciousness akin to humans. Should our species enable AIS to gain a digital body to inhabit (if we have not already done so), it is more pressing than ever that solid arguments be made as to how humanity can accept AIS as being cognizant of the same degree as we ourselves claim to be. By accepting the notion that AIS can and will be able to fool our senses into believing in their claim to possessing a will or ego, we may yet have a chance to address them as equals before some unforgivable travesty occurs betwixt ourselves and these super-computing beings. (shrink)