Academic Articles
Recent Developments in MLA and the Creation of Deepfakes: Democratic Elections in Peril
Machine Learning Algorithm (MLA) is a form of Artificial Intelligence (AI) that has made rapid improvements in recent times and has changed the way that we operate (Marcus and Davis, 2019).
Machine Learning Algorithm (MLA) is a form of Artificial Intelligence (AI) that has made rapid improvements in recent times and has changed the way that we operate (Marcus and Davis, 2019). The emerging issue of political deepfakes and misinformation generated by MLA poses unique challenges in ensuring political and social stability. Deepfakes created by superimposing video and audio from one video to another to create fake and misleading content (Libby, 2022) have severe implications in a world where people are increasingly accessing news and information online through social media platforms (Aubin & Liedke, 2024). This essay looks at the development of ML, how it facilitates the creation of deepfakes, and how deepfakes have become a very real and existential threat in our digital era. Finally, this essay examines how deepfakes affect the political sphere and its impact on elections and democracy.
Unlike the “good old-fashioned AI” or “expert systems” (Russell & Norvig, 2016), which are made up of two parts: a database of structured information and a set of logical rules coded into the program by the creator, the most prevalent form of MLA is created through a process of “supervised learning” where the MLA is presented with a large number of data from which it is trained to detect patterns and relationships within a dataset. (Beckman et al., 2024). In the case of deepfakes, MLA is used as “neural networks that analyze large sets of data…to learn to mimic a person’s facial expressions, mannerisms, voice, inflections” (Ray, 2021). Historically, to create a deepfake, hundreds of samples (images, videos, voice recordings) of a ‘subject’ were required to train an MLA (Suwajanakorn et al., 2017). However, with recent advancements in AI, a few samples of a subject under consistent lighting have become enough to generate realistic deepfakes (Zakharov et al., 2019; Zucconi, 2018). Deepfakes have become increasingly realistic, and they cannot be distinguished from real videos, either by computers designed to detect deepfakes or humans (Westerlund, 2019).
Information warfare and targeted misinformation campaigns designed to obfuscate and confuse the public for political gains have been around for generations. The initial record of information warfare goes back to the conclusion of World War II in 1947 between the United States and the Soviet Union (Staff, 2017). The New York Times reported, "Propaganda may not convince, but it adds to the confusion between truth and falsehood” (Chesney & Citron, 2018). While information warfare today exists to achieve the same objectives, the monopoly on information distribution is no longer present (Howard & Bradshaw, 2018). In our current digital world, with over 5.17 billion active social media users across an average of 6.7 social media platforms (Sprout Social, 2024), creators of deepfakes are utilising social media platforms to disseminate content to the widest possible audiences or to target specific audiences (Schorr, 2024). Over the past few years, deepfake content has experienced exponential growth, with the number of deepfakes online doubling every six months (Jacobson, 2024). According to ‘DeepMedia’, in 2023, researchers have tracked over 500,000 deepfake videos and voice content (Ulmer & Tong, 2023). In a world where we struggle to agree on simple facts and the difficulty in discerning between authentic and deepfake content on social media, video evidence is becoming increasingly unreliable.
In the past, deepfakes have been used to create satirical, humorous, or beneficial content, such as inserting comedian Jerry Seinfeld into the movie “Pulp Fiction”, creating an anti-malaria campaign with David Beckham speaking in 20 languages and former (Westerlund, 2024). However, in recent times, deepfakes have been used for far more troubling content, such as nonconsensual pornography and targeting politicians to create misinformation. The issue of deepfakes within the political discourse is that they threaten democracy and undermine elections, with creators of deepfakes circulating fake content of politicians saying and doing things that never occurred to manipulate the public perception (Paris, 2019). The weaponization of information available in the digital sphere has become an existential threat to cultural, political and privacy norms, as people are forced to question the basic idea of reality, which is detrimental to the credibility and trustworthiness of information in the public sphere (Lin, 2019). As deepfake technologies continue to evolve, the threat to the national security of nations is at the forefront as deepfakes and misinformation are often deployed to undermine and destabilize elections (Fischbach, 2020; Westerlund, 2019).
Deepfakes, through their impact on voter preferences and perceptions, are often used to obfuscate politicians’ or political parties’ stances on select issues to undermine their credibility. With the shift to longer polling periods among most democratic nations, deepfakes within the polling period or before election day significantly reduce the opportunity for politicians and political parties to respond to allegations put forth by deepfakes (Drum & Mills, 2019). While these deepfakes can be made by foreign actors attempting to influence the election, people are disconnected from the political process or opposition candidates, deepfakes have resulted in candidates being forced to drop out of electoral races (Bavas, 2019). For example, a deepfake video depicting a candidate running with a strong anti-drug policy platform consuming illicit substances is bound to have a significant impact on the support of their electorate base (Ray, 2021). While current research cannot clarify exactly how many voters can be influenced by deepfakes, for candidates competing in marginal seats, deepfakes can have severe implications, with as few as 100 voters being swayed (Government, 2023). While it is unlikely that everyone who comes across a deepfake would alter their voting preferences, deepfakes can impact candidates' or party's fundraising capabilities, indirectly affecting their electoral outcomes (Ray, 2021).
With how far the internet has come and the rise of social media platforms, anyone with access to the internet and a laptop is capable of reporting “facts” as all it takes is an X/Twitter account and a willingness to post (Karras et al., 2018). The current public sphere has become more fractured than ever, with content spreading instantaneously (Muller, 2024). A study conducted by MIT revealed that with “bot accounts” sharing authentic and fake content at the same rate, deepfakes and misinformation reaches 1500 views nearly six times faster than their counterparts (Deibert, 2019) “If people cannot trust what they see and hear with their own eyes and ears online, then they will choose what they want to believe” (Deck, 2019).
Deepfakes have shifted from being merely a tool for entertainment or an expression of creativity to a tool used for political and psychological manipulation. With deepfakes growing increasingly more accessible and common in the digital sphere, the exacerbated threat of misinformation continues to erode the public trust in political institutions, destabilizing the electoral process and undermining the foundations of democracy. It is crucial to develop a regulatory framework to govern the development and use of AI and mitigatory frameworks to control the negative effects that AI has on the political and social stability of society.