Artificial Intelligence in the World: Mapping its uses during the 2024 elections

By: FGV Comunicação Rio

Executive Summary

• Although the pre-election period was marked by high expectations for the extensive use of AI in the electoral process, this did not materialize in practice. Whether due to the pioneering actions of the Superior Electoral Court (TSE), which approved a series of regulations governing AI use in elections, or due to the specific characteristics of municipal elections, the use of AI technologies in the 2024 campaigns was not widespread and was limited to more specific applications; 
• In Brazilian municipal elections, the predominant challenges related to AI use are not necessarily connected to misinformation. While the potential to amplify disinformation narratives remains a key concern and a frequently debated topic, during the municipal elections, AI applications were not heavily linked to informational issues, as such content is often easily recognized as edited. Instead, we identified various uses focused on political critique, demands, and attacks; 
• Deepfakes embrace the "jeitinho brasileiro" and are strongly tied to humor. Many of the videos and images identified during the campaign exhibited a playful or humorous nature, often adopting a non-professional, meme-like aesthetic. These contents often mocked or ridiculed political figures, leveraging AI tools to produce and share such materials; 
• Preexisting social issues and dynamics find new avenues through AI. One of the most notable topics in the elections involved the creation of deepfakes aimed at political violence against female candidates. While the fraudulent nature of many such contents could be easily recognized, their dissemination highlighted how AI amplifies methods of ridiculing, shaming, exposing, and silencing women in politics;
• Candidates and public institutions are investing in AI to enhance citizenship. During the 2024 elections, the use of chatbots to provide informational support to citizens and voters stood out, reinforcing the notion that AI, in and of itself, should not be viewed as harmful to democracy; 
• When asked about the elections, generative AIs struggle with accuracy and source selection. An analysis of the models Copilot, ChatGPT 4o and 4o-mini, Gemini, and Claude 3.5 Sonnet revealed variations in the quality and accuracy of their responses regarding Brazilian elections, including inconsistencies about political figures and incorrect information about the electoral system. Examples of such inaccuracies include the fabrication of a Harvard master's degree for Eduardo Paes (PSD-RJ) and the erroneous claim that Ricardo Nunes (MDB-SP) was still São Paulo's vice-mayor.  

Access the full report here.