Artificial Intelligence in the world

By: FGV ECMI

 

Executive Summary

• Deepfakes are the most visible dimension of AI in elections: The use of Artificial Intelligence (AI) to create manipulated videos, audios, and images is present in all the elections analyzed. The empirical cases demonstrate a wide range of applications, from the creation of explicitly disinformative content to the use of historical figures to bolster political support for candidates. Beyond the diversity and adaptation to the specific contexts of each election, as detailed throughout the study, it is important to note that the origin of these contents is often uncertain and is sometimes directly or allegedly linked to parties, candidates, and their allies;
• AI-generated political content aligns with local contexts: Despite the widespread use of AI around the world and the transnational nature of disinformation strategies, the cases analyzed highlight the importance of local contexts in the mobilization of electoral narratives. These narratives include those aimed at attacks and disinformation against political opponents, as well as strategies intended to strengthen candidates and political parties. This dimension is evident in the themes utilized in AI-generated content. For instance, in India, religion plays a crucial role in the electoral debate and is reflected in AI usage. In Mexico, the theme of violence is also mobilized in deepfakes and AI-powered content. Additionally, deepfakes simulating videos of public figures like Trump and Eminem criticizing Nelson Mandela's party in a polarized election in South Africa further illustrate the significance of national contexts in the use of AI as a political strategy in elections worldwide;
• Chatbots balance between the accuracy of information and content filters. A less visible aspect of the growing use of generative AI tools is the automated creation of text through tools like chatbots. To assess the accuracy of these tools – Copilot Criativo, ChatGPT (3.5 and 4.0), and Gemini – a standardized questionnaire with both informative and opinion-based questions about the elections in question was applied. The analysis highlights that while these AIs provide mostly correct data about electoral contexts and voting systems, they often present outdated information and vary in their responses. Additionally, they heavily rely on the content filters designed by the respective companies, underscoring the ongoing need for regulation and control to ensure the integrity of electoral information and address the selection of topics that chatbots do not respond to;
• Institutional regulation is already part of the debate, with the European Union taking the lead, while legislative gaps remain evident: The regulatory component regarding the circulation of political content on the internet and the use of AI technologies in electoral contexts is a consistent theme across the cases studied. In the European Union, this is the central element of the debate, not only due to the existence of regulations concerning the online circulation of political content but also due to the recent approval of a normative framework for Artificial Intelligence. While this context has encouraged the signing of voluntary agreements between parties in the European Parliament elections, resulting in a more advanced and constructive debate on the issue, it has not completely prevented the circulation of deepfakes, occasionally appearing on the pages of national parties. In other countries, emerging debates and a noticeable absence of specific instruments were observed.

Elections in India: The Use of Deepfakes to "Resurrect" Political Support in the World's Largest Election

The electoral campaign in India was not only the largest in the world in terms of electorate but also stood out for its strategic use of emerging technologies like AI. Deepfakes populated the election that led to Narendra Modi's re-election as Prime Minister. AI-generated content was allegedly used by parties and candidates to spread false displays of support from celebrities to even "messages from beyond," featuring visual reproductions of deceased political figures. In a communication landscape characterized by low internet penetration alongside a large number of digital platform users, the large-scale use of AI had a significant impact on the election, drawing the attention of global observers and raising alarms about the risks of this technology in disseminating misinformation in the country.

The electoral dispute in India is characterized by economic issues, threats to democracy, and religious conflicts that mark the elections.

Uses of AI: Deepfakes of Deceased Celebrities and Politicians in the Indian Election

In the Indian elections, there is a widespread use of Artificial Intelligence. Beyond citizens, political parties, media agencies, campaigns, and candidates utilize AI to promote ideas and facilitate political and electoral actions in a significant way. This acceptance of technology is evident in how political actors, such as Narendra Modi himself, use AI with the aim of promoting linguistic inclusion in the country. Political parties and official campaigns are also associated with the dissemination of deepfake political content in the 2024 elections, resulting in curious uses and risks related to misinformation and negative campaigning.

The widespread use of AI-generated videos during the 2024 elections in India even involved deceased political figures, whose visual and vocal representations were used to express support for certain political factions. One example is the former Chief Minister of Tamil Nadu, J. Jayalalithaa, whose voice was manipulated in content critical of the party currently governing the state. 

 

Elections in South Africa: Deepfakes are used to misinform in a contested election

Two aspects stood out at the intersection of elections and Artificial Intelligence in South Africa. In the highly competitive 2024 elections, where the African National Congress (ANC), the historic party of Nelson Mandela, faced its biggest electoral challenge in three decades of governance, the dissemination of political content prominently featured the use of AI, including direct posts from the daughter of Jacob Zuma on X, the former president and main opposition leader.

Deepfake cases were identified in attempts to influence public opinion by mobilizing support for political candidates opposing the ANC. The fabricated videos promoted the idea of foreign political support from figures like singer Eminem and former President Donald Trump, who appeared to criticize Mandela's party and endorse opposition candidacies.

Electoral contest in South Africa: Mandela's party faces public scrutiny after 30 years in power

 

Predominating the use of deepfakes over other AI-based approaches, there is a tendency in the country to use this technique to simulate political opinions and positions, usually against the ANC. Additionally, other uses, such as the forgery of documents and signatures through artificial intelligence software, can be highlighted, but incidents of deepfakes gain much more significance in the South African context.

Elections in Mexico:  Diversified uses of AIs characterize an election marked by a wave of political violence.

In 2024, Mexico witnessed the election with the highest voter turnout in its modern history, electing the first female president of the country. While the election was the most significant in terms of volume and one of the most symbolic regarding the advancement of female participation in politics, it was also the most violent. From June 2023 until the election, around 38 candidates were assassinated, and 560 candidates and electoral officials received government security due to a history of threats.

The electoral dispute in Mexico: a wave of violence and crime overshadows the victory of the country’s first female president

 

Interestingly, even amid the electoral scenario, forgeries aimed at the economic and financial sectors draw attention, having no direct relations to the election. That is, these are contents aimed at fraud that utilize images of political figures but do not mention any aspects related to the elections.. 

This is the case with the video of President Andrés Manuel López Obrador inviting citizens to invest in shares of the oil company Pemex, for example, or the content featuring López Obrador and presenter Joaquín López-Dóriga promoting an investment app designed by Elon Musk. The latter was verified by Reuters after being shared more than 2,000 times on Facebook. This type of manipulation of images, audio, and videos also targeted then-presidential candidate Claudia Sheinbaum. During the campaign, some content showed Sheinbaum promoting a supposedly miraculous financial investment circulating on social media, with its veracity also challenged by Reuters. 

 

Elections in the European Parliament:  

Deepfakes still circulate in an election where the regulatory component is highlighted

In the context of the 2024 European Parliament elections, the intersection of digital regulation and the electoral environment emerged as a central theme, reflecting both the complexities of the European political landscape and the growing role of technology in politics. With a population of approximately 448.4 million voters and a robust digital environment, the European Union experienced a vote marked by the influence of new regulatory norms, including the recent implementation of the Digital Services Act and the approval of the Artificial Intelligence Act within the bloc. The election results indicated a rise in right-wing parties compared to the last legislature, while the leadership of the EPP-S&D-Renew coalition at the center remained intact, allowing for the re-election of Ursula von der Leyen as President of the European Commission, a figure directly involved in the digital agenda adopted by the bloc in recent years.

This regulatory context not only distinguishes the European Parliament elections from the other cases analyzed, with characteristics such as the voluntary signing of commitment agreements among political parties and the low incidence of misuse of technology in the electoral competition, but it also frames the debate in more mature terms, revealing still-present contradictions. This is because the circulation of AI-generated content, again with the technology of deepfakes, was nevertheless part of the electoral context of the bloc

Regulation fosters a proactive environment, highlighting the role of political parties in the EU election

National elections mobilize debate on the use of AI in Europe

In France, Loïc Signor, a leader of the party founded by Emmanuel Macron, Renaissance, published a deepfake showing Marine Le Pen speaking in Russian in the months leading up to the European elections. When questioned, Signor indicated that, in addition to denouncing Le Pen's alleged ties to Russia, the deepfake was published to raise awareness of the risks associated with artificial intelligence. 

Still in France, an analysis of the main political parties' social media pages between May 1 and June 28 mapped various posts with images generated by AI, with a strong presence among the far-right parties Rassemblement National (RN), Reconquête, and Les Patriotes. The content of these posts mapped associations with rhetoric of fear and moral panic, reinforcing a nationalist tone, whether in anti-immigration posts or in images pointing to a threat to freedom of expression in the country. 

 

Other uses of AI in the electoral context: what do generative AI models have to say about the elections?

The presented cases of AI usage demonstrate how technologies, particularly Generative Artificial Intelligence (GAI), have been concretely present in electoral processes around the world, indicating the transnational circulation of pieces created with deepfake technology. Another prominent generative artificial intelligence technology today, whose mapping of use in the surveyed elections is less identifiable, involves models currently focused on text generation based on commands sent by users of tools like OpenAI's ChatGPT, such as questions regarding, for example, the electoral processes themselves, thus serving as a source of consultation for voters and technology for the rapid creation of texts. 

AI companies invest in adaptations of technology for election

Disinformation cases through AIs have raised alarms among governments worldwide. In response to these concerns, leading companies in the sector have published statements on their websites outlining actions to prevent the harmful use of their technologies in elections. For tech giants, this type of publication is already traditional on official blogs, which often outline the strategies and concerns of these organizations as a way to signal to governments and the market.

Since the end of 2023, the three tech giants with consumer AI models (Google, Meta, and Microsoft) have published their policies regarding the use of their AIs during the electoral period this year in nearly 80 countries, mobilizing an electorate of more than 4 billion people. On the startup side, OpenAI, the developer of ChatGPT and DALL·E, and Anthropic, creator of Claude, have also made publications, but their image-focused counterparts, like Midjourney and StabilityAI, did not follow the trend. For this reason, each of these elections tends to be a small experimental laboratory of the different uses that artificial intelligence tools can have in electoral campaigns and the control mechanisms that democracies can mobilize to prevent these attacks.

However, months before the research was conducted, Google decided to block Gemini from providing answers about ongoing elections. These restrictions aim to limit the influence that the model may have on manipulating public opinion.

Expediente 

Diretor

Marco Aurelio Ruediger

E-mail: marco.ruediger@fgv.br

Vice-diretor

Amaro Grassi

E-mail: amaro.grassi@fgv.br

Pesquisadores 

Letícia Sabbatini

Laura Pereira

Leonor Jungstedt

Victor Piaia

Mariana Carvalho

Catherine Moura

Mario Lima

Alessandra Maia

Projeto gráfico

Daniel Almada

Luis Gomes

Access the full report here.