Deepfakes in the 2020 Presidential Election: Detecting New Attempts at Deception

Session Submission Type: Full Paper Panel

Session Description

This panel examines the growing problem of deepfake videos and disinformation in the context of the 2020 US presidential election. While the 2016 election cycle was characterized by an unprecedented rise of digital disinformation, much of it conveyed through porous social media platforms without much user awareness of its origins, the 2020 election has witnessed the rise of the next generation of informational distortion in the form of manipulated videos of politicians that appear real but aren’t. So-called “deepfake” videos, made possible by advances in artificial intelligence and video editing software that allows existing footage of a target candidate or politician to be synthesized and altered with someone else's likeness, mannerisms, and speech, are creating opportunities for propagandists to further imperil the already compromised political communication environment by calling trust in professional journalism and the wider system of public information on which democracy rests itself into question. Without confidence in the informational foundation of politics and policy making, all arguments and assertions are equally valid—and equally suspect.

The vulnerabilities that deepfakes exploit operate at both the organizational and personal level. Organizationally, the 24-hour news cycle and hyper-commercialization of US media places a premium on scooping the competition and breaking new developments first, often at the expense of accuracy. In this rush to publish, false information can circulate before it is identified as fraudulent. This problem is even more pressing on digital platforms. In the wake of the 2016 election, several studies documented how fake news often spreads faster than real news on social media. More recently, grossly distorted videos of journalists and Democratic members of Congress, including Jim Acosta of CNN and House Speaker Nancy Pelosi, were circulated for political purposes and conveyed by some as real. Although Twitter and Facebook have banned deepfake videos ahead of the 2020 presidential election, encompassing content that has been edited “in ways that aren’t apparent to an average person and would likely mislead someone” if viewed, this does not prevent circulation of videos that are considered parody or satire—or footage that has been edited to remove words or change the order in which statements occur. Thus, further distortions are not just likely but inevitable.

On a personal level, deepfakes pose a serious concern because the typical voter does not have the interest, technical skill or motivation to research whether a given video presented as real is bunk or actually holds informational currency. As psychological research on “believing the unbelievable” has shown, the tendency when confronted with new information is to accept it as true first and then only reject it subsequently if assessed as false. In most situations, the deck is stacked in favor of accepting false information rather than rejecting it—a proposition even more likely to happen when a deepfake video appears indistinguishable from the real thing. Recent research by Vaccari and Chadwick (2020) has documented how just under 50% of participants in a study of a deepfake video posted for educational purposes by Buzzfeed of former President Obama were deceived by it, despite the video’s improbability (e.g., Obama publicly calling Donald Trump a “dipshit” in a televised video). They conclude that people are more likely to feel uncertain than to be misled by deepfakes, but this resulting uncertainty, in turn, can reduce trust in news, particularly on social media. Deepfakes may thus contribute toward generalized indeterminacy and political cynicism, further intensifying recent challenges to online civic culture in democratic societies.

In the face of these challenges and developments, this panel presents four original studies that examine the deepfake threat from both social scientific and computational perspectives. Papers in the panel utilize a variety of methodologies to focus on the challenge of deepfake detection from both a human and algorithmic point of view—and the consequences of exposure to these new forms of visual propaganda to viewers. The panel consists of an international and interdisciplinary group of scholars from political science, communication research, and information science, with a mix of ranks at the junior and senior level. The panel also features two highly respected discussants: Andy Chadwick of Loughborough University in the UK and Robert X. Browning of Purdue University in the US.

Sub Unit

• Related Groups / Association for Politics and the Life Sciences

Individual Presentations

• Creating, Weaponizing, and Detecting Deep Fakes - Shruti Agarwal

• Do (Microtargeted) Deepfakes Have Real Effects on Political Attitudes? - Claes H. de Vreese, University of Amsterdam; Tom Dobber, Universiteit van Amsterdam

• The Manipulation Effect: How Small Editing Changes to Political Video Influence Viewer Perception - Erik P. Bucy, Texas Tech University; Zijian Harrison Gong, Texas tech; Jungseock Joo, UCLA

• When Are Deep Fakes Effective? Perceptual Fluency and Subtle Digital Alterations - Bryce Dietrich, University of Iowa

Chair

• Erik P. Bucy Texas Tech University

Discussants

• Andrew Chadwick Loughborough University

• Robert X. Browning Purdue University