Deep Fakes

Algorithms and Society

Michael Filimowicz

• Publisher: Routledge
• Year: 2022
• Print ISBN: 9781032002620
• eBook ISBN: 9781000588774
• Edition: 1
• Topic: Social Sciences
• Subtopic: Sociology
• Index: Social Sciences

Deep Fakes: Algorithms and Society focuses on the use of artificial intelligence technologies to produce fictitious photorealistic audiovisual clips that are indistinguishable from traditional video media.
For over a century, the indexical relationship of the photographic image, and its related media of film and video, to the scene of capture has served as a basis for truth claims. Historically, the iconicity of these images has featured a causal traceback to actual light rays in a particular time and space, which were fixed by chemical reactions or digital sensors to the resultant image. Today, photorealistic audiovisual media can be generated from deep learning networks that sever any connection to an actual event. Should society instantiate new regimes to manage this new challenge to our sense of reality and the traditional evidential capacities of the 'mechanical image'? How do these images generate information disorder while also providing the basis for legitimate tools used in entertainment and creative industries?
Scholars and students from many backgrounds, as well as policymakers, journalists and the general reading public, will find a multidisciplinary approach to questions posed by deep fake research from Communication, International Studies, Writing and Rhetoric.

1Deep Fakes Seeing and Not Believing

Sean Maher

DOI: 10.4324/9781003173397-1

Introduction

Deep fakes describe videos that employ artificial intelligence (AI) technology and represent false and misleading media depicting events that never occurred. By compounding the phenomenon of fake news, deep fakes underline how journalism, news and information in the digital era can be compromised. Despite being a relatively recent phenomenon, deep fakes join traditions in media representations that include hoaxes and fabrications designed as deliberate ploys to mislead and distort legitimate news and information services. Since July 2019, the specific term “deep fake” has gained currency and in conjunction with fake news constitutes part of the broader problem arising from “information disorder”. Symptomatic of what has been referred to as a post-facts era, information disorder describes the effects of notions like the relativity of truth and the threats this represents to an informed citizenry and functioning polity (Anderson, 2018; Qayyum et al., 2019; Westerlund, 2019), Zannettou et al. (2019).

Deep fakes, also referred to as synthetic videos, can infiltrate news and media organizations from social media platforms through deliberate and unintended distribution. As media products, deep fakes result from layered manipulations of previous datasets harvested from video and audio sources comprising a spectrum of sound- and image-based deceptions. To understand deep fakes and differentiate them from shallow fakes, consideration needs to be given to how they are produced and the role they play in creating confusion surrounding trusted news sources.

Trust in factual data as well as the role of public institutions is connected to reliable news and information sources. Deep fakes compound the corrosive force of fake news, a term more recently conflated with criticism of legitimate information and journalism out of personal or political self-interest. Information disorder is a more encompassing notion that covers how sophisticated and malicious actors employ “computational amplification” of fabricated news and data. Functioning as dis-information that manifests as deep fakes to fake social media profiles, dis-information operates by manipulating online content as well as search engine results. In a 2017 Council of Europe report, dis-information is outlined as “sowing mistrust and confusion and to sharpen existing socio-cultural divisions using nationalistic, ethnic, racial and religious tensions” (Wardle & Derakhshan, 2017, p. 1). Responding to deep fakes means locating their effects within the context of dis-information and understanding their role in compromising legitimate, fact-based information and authorized knowledge systems.

As defined by Wardle and Derakhshan, this discussion approaches deep fakes in the context of what distinguishes mis-information, dis-information and mal-information:

• Mis-information is when false information is shared, but no harm is meant.
• Dis-information is when false information is knowingly shared to cause harm.
• Mal-information is when genuine information is shared to cause harm, often by moving information designed to stay private into the public sphere.
  (Wardle & Derakhshan, 2017)

By canvassing the critical response from communications scholarship, the discussion maps deep fakes in the context of information disorder and the measures being taken to ameliorate them in the evolving digital landscape and media ecology. To combat the harmful effects of information disorder, a key strategy involves the fostering of institutional resiliency, something that continues to gain currency and momentum. Notwithstanding the damaging effects deep fakes can have on specific individuals involved in their distorted representations, it is in the broader journalistic and communication domains of news and information where deep fakes may have some of the most profound effects on the broader social polity. Deep fakes have renewed calls for reforms across education and policymaking that is being marshaled to help withstand the onslaught of information disorder. By outlining these developments the discussion situates the emergence of deep fakes against the many opportunities and challenges from wider social transformations from the digital economy and information technology.

A New Media Ecology

Polity and the public space of political thought have been redefined by a transforming communication landscape challenging the institutions of journalism and formal modes of authorized information. Disseminated from a top-down public sphere, the traditionally sober rationality of news reflected the command-and-control nature of information from mass media communications. In what is referred to as legacy media, publicly distributed information was primarily managed within the confines of a media establishment with expert communication professionals and reputable news organizations. In contrast, informal, conversational, opinion-centered communications that are often emotion-laden circulated in relatively transparent and discrete domains as nonprofessional discourse.

As the new space of media and communications continues to take shape, it is clear the former distinct spheres of communication once divided along public and private, professional and amateur, conversational and informational, rational and emotional have collapsed. Among the many transformations generated by web-based media, one of the key influences has come about from what has been described as “spreadable media” (Jenkins et al., 2013) and the rise of the amateur (Flichy, 2007). Facilitated by the ubiquity of digital affordances and propelled by social media platforms, the distributive impact of the amateur/user-generated continues to alter social, political and cultural dimensions of everyday and institutional life in the twenty-first century. The many forms of actions and interactions characteristic of online behaviors hearken to what Castells identified as the operations of “network society” – “a society whose social structure is made up of networks powered by micro-electronics-based information and communications technologies” (Castells, 2004, p. 3). The repercussions from network society on democracy are interpreted by Ferreira as “a shift from the notion of participatory democracy to that of a cooperative democracy, built ‘from below’ by a public that speaks without asking anybody” (Ferreira, 2016). The democratizing benefits of these shifts in communication are accompanied by the many challenges issuing from the attendant “digital disruptions” and unresolved challenges surrounding the creation and distribution of reliable information. The disintermediation of transforming supply lines of commercial activities underpinning the business models of e-commerce sees either wholesale suppliers directly linked to consumers or the number of agents between consumers and producers substantially reduced. In communication domains, however, processes of disintermediation have reduced or eliminated the role performed by publishers and news organizations as gatekeepers that contain the spread of disinformation.

As systems and processes governing the business of communications are irrevocably ceded from legacy media institutions, journalism and reliable information continue to be reconfigured against a new mediascape. While always contentious and imperfect in their relationship to news and journalism, traditional media organizations were fundamentally guided by principles of transparency, balance and accountability. As formal publishers of information, it was incumbent upon media organizations to uphold verification and authentication measures that saw them function in terms of a quality assurance mechanism. By rigorously checking the veracity of the sources of information, levels of public confidence were met by the obligations of responsible reportage that safeguarded against inauthentic news and promoted ethical responsibility.

In contrast, the phenomenon of false news coincident with the rise of social media illustrates the distributive impact of web-based and peer-to-peer-style communications. Legally constructed on different terms from publishers since the 1998 Digital Millennium Copyright Act (DMCA), internet service providers and intermediaries like social media platforms are afforded protections from liability on the legal basis of functioning as neutral hosts of user content. The marked legal and operational differences between social media platforms originating in the tech sector from traditional publishers mean operators like Facebook and Google have no editorial remit and have traditionally resisted takedown orders of user-generated content. The online media context also means the scale of content creation and those producing it have grown exponentially. Founded in 2004, in 2021 Facebook has grown to 2.74 billion monthly active users. According to Newberry, between 2014 and 2020, across Australia, Brazil, Canada, Denmark, France, Germany, Ireland, Italy, Japan, the United Kingdom and the United States, 36% of people received their news from Facebook (Newberry, 2021). Meanwhile, Google searches are estimated to be between 3.5 and 5.6 billion per day, which represents a 90% market share and provides parent company, Alphabet, with 80% of its revenue from advertising, which in 2019 totaled $US147 billion (Desjardins, 2018) and (Graham & Elias, 2021).

Reconfigured by web-based communications, the traditional media establishment and its legacy institutions not only have had to compete with firms from a rising tech sector but have also become increasingly vulnerable to being enlisted in the distribution of false and misleading news and information. Attributed to the “platformization” of newsrooms, diminished editorial oversight compromises balanced reporting. It combines with “financialization and metricization [to] push journalists and editors to produce more content, faster, and with fewer resources” (Donovan & Boyd, 2021, p. 339). The commercial dictates and 24/7 news cycle place a greater premium than ever before on “being first” and ensure traditional media organizations are forever chasing breaking news that compromises their ability to provide gatekeeping on news stories. According to Villi, “social media is the third most preferred gateway to online news, behind direct access to media outlets and almost on par with online searches” (Villi, 2019, p. 2). Traditional news media organizations have had to respond with automated social listening tools that pick up on trending stories and means of republishing them without editorial-based fact-checking.

Social media platforms and user-generated content mean the contemporary media ecology is characterized by online distribution systems that fragment news and information and diminish editorial oversight and organizational accountability. Presenting some unprecedented challenges for the global digital mediascape, the arrival of deep fakes has emphasized the need for a more concerted and robust response to information disorder. The proliferation of false news items and misleading information that manages to circulate freely online with relative impunity requires concerted intervention. There is no underestimating the scale of the challenge given that since its arrival the internet has been promoting fierce resistance at any efforts to regulate it. From tech giants to operators and users of digital currencies like Ethereum and Bitcoin, the online world still holds out utopian promises based on decentralization and the ending of state and corporate monopolies that are free of regulation. Striking a balance to support opportunities posed by the new and emerging frontiers of the virtual world with the assurances accompanying policies and enforcement of good governance remains.

Mining Deep Fakes

Face-swap videos enabled by AI that became known as deep fakes emerged as recently as 2017. Representing a relatively new phenomenon to the suite of creations that target mis-information campaigns, deep fakes are enabled by the machine learning of AI technology. But like the many forms of fake and fraudulent communications preceding them, deep fakes are characterized by anonymous creators and the ability to capitalize on automated forms of distribution across web-based communication systems. Despite posing obvious risks, deep fake technology was ostensibly developed with benign intentions that sought to utilize large datasets of available footage that was for all intents and purposes intended for authorized usage. The AI technology beneath deep fakes is employed in legitimate applications in the film and gaming industries and can extend the kinds of virtual reality experiences sought by museums and educational media. In 2019, for example, deep fake technology facilitated an interactive experience that enabled guests at the Dali Museum in St. Petersburg, Florida, to appear to have a personal encounter with Salvadore Dali (theverge.com, Lee, May 10, 2019). Entertainment and gaming industries continue to develop and employ the technology for audio dubbing, special effects and low-cost production values that can be used to enhance high-budget and low-cost amateur productions alike.

The creation of deep fake videos requires a large dataset of publicly available video footage and audio files, then employing deep learning achieved by machines teaching themselves as machine learning models. Deep fakes are made possible by two neural networks known as generative adversarial networks (GAN) working together as the “generator” and the “discriminator”. The first machine learning process, the generator, involves training on a dataset composed of a large swathe of images drawn from publicly available sites that enables it to learn and mimic a person’s facial expressions and voice. The harvested images center on a target subject that includes as many face and body angles with features and expressions under multiple lighting conditions. Video forgeries are created from the data collected in this first stage of the process after swapping it onto another person by employing a deep learning algorithm. Once a video ...

Table of contents

• Cover
• Half Title
• Series
• Title
• Copyright
• Contents
• List of Contributors
• Series Preface: Algorithms and Society
• Volume Introduction
• 1 Deep Fakes: Seeing and Not Believing
• 2 Deep fakes and Disinformation in Asia
• 3 On the Depth of Fakeness
• Index