The Phantom Menace That Wasn’t: AI’s Limited Impact on Recent Elections
The 2024 election cycle, both in the United States and abroad, unfolded under the looming shadow of a new technological threat: generative artificial intelligence (GenAI). The potential for AI-powered deepfakes – convincingly realistic fabricated audio and video content – to manipulate public opinion and disrupt democratic processes sparked widespread concern. Experts and commentators predicted a deluge of AI-generated disinformation, blurring the lines between truth and falsehood and eroding public trust. However, post-election analyses paint a different picture. While AI-generated content did appear, its impact was far less significant than anticipated, failing to live up to the pre-election hype.
Studies conducted by organizations like the Alan Turing Institute’s Centre for Technology and Security (CETaS) reveal a surprisingly limited use of AI-generated disinformation in elections across the UK, France, the European Union, and the United States. These campaigns, few in number and often amateurish in execution, reached only small, niche audiences whose political views already aligned with the disseminated narratives. The CETaS research concluded that AI-driven disinformation had no measurable impact on election outcomes in these regions, primarily serving to reinforce pre-existing beliefs rather than swaying undecided voters. This limited impact is further corroborated by low exposure rates among the general public. In the UK, for instance, only a small fraction of survey respondents reported encountering politically motivated deepfakes, despite widespread awareness and concern about the technology.
The U.S. election landscape, often seen as a prime target for disinformation campaigns, similarly saw a limited role for AI-generated content. Analysis of fake news circulating during the election cycle revealed that only a small percentage was produced using GenAI. Major tech companies, including Microsoft, Meta, Google, and OpenAI, reported minimal distribution of AI-generated content and a lack of substantial foreign interference leveraging AI. This suggests that the anticipated wave of sophisticated, AI-driven influence operations failed to materialize. Instead, much of the observed AI content was crude and easily identifiable, suggesting amateur creation rather than orchestrated campaigns by sophisticated actors.
Interestingly, public discussion surrounding AI and elections often focused more on the release of new AI models than on actual election events. Analysis of social media mentions of terms like "deepfake" and "AI-generated" showed spikes coinciding with the launch of tools like ChatGPT-4 and Grok, while mentions during key election dates remained relatively low. This pattern highlights the disproportionate attention given to the potential of AI manipulation compared to its actual observed use in electoral contexts. The trend of limited AI-generated content and minimal impact extended beyond Western democracies. Elections in countries like Bangladesh and South Africa saw a similar scarcity of AI-driven disinformation, further reinforcing the global picture of AI’s muted role in recent electoral processes.
The content that did emerge was often easily identifiable as AI-generated, frequently carrying telltale signs of the tools used in its creation. Examples include a deepfake image of Kamala Harris with Stalin, a fake video of her addressing a communist rally, and an AI-generated image protesting insect-based diets. These examples, often circulated within partisan echo chambers, underscore the amateur nature of much of the AI-generated political content. Furthermore, a significant portion of AI-generated content during these elections was intended for satire or entertainment, rather than malicious manipulation. This further diminishes the narrative of AI as a primary tool for electoral disinformation.
While the direct impact of AI-generated content remained limited, the mere awareness of its potential had a more pronounced effect. The fear of AI-driven manipulation contributed to a general climate of distrust in online information. Misinformation, regardless of its origin, became readily attributed to AI, further blurring the line between authentic and fabricated content. This phenomenon, observed in social media discussions, highlights the broader challenge posed by AI to information integrity, even in the absence of widespread malicious use. The "cry wolf" effect, whereby genuine content is dismissed as AI-generated, raises serious concerns about the erosion of trust and the potential for chilling legitimate political discourse.
The findings from the recent election cycles suggest that while the prevalence and demonstrable impact of AI-generated content remained negligible, the mere awareness of its potential has created a significant impact on public perception and trust. This awareness, coupled with the difficulty many individuals face in discerning real from fake content online, has created a fertile ground for skepticism and distrust, potentially undermining faith in democratic processes. While the feared flood of sophisticated deepfakes failed to materialize, the very possibility of their existence has cast a long shadow over the information landscape. The challenge moving forward lies in addressing this underlying distrust and promoting media literacy to mitigate the broader implications of AI for democratic discourse and public trust.
