KI/Demokratie-Initiative

Project Methodology: A Data-Driven Approach 

To draw conclusions about the role AI played in India, South Africa, France, Mexico, Germany, and the United States, the first step involved was to identify AI incidents using a data-driven, quantitative approach. The primary aim of this analysis was to examine how frequently AI incidents occurred during elections and the nature of the threats they posed, if they did. These insights were used to identify trends and vulnerabilities that countries faced in 2024 concerning AI and elections. 

One of the central challenges of this data-driven approach lies in the nondisclosure of AI use in election campaigns. Political actors often lack incentives to reveal their use of AI technologies, and the absence of robust platform regulations mandating the labeling of AI-generated content further complicates reliable identification. These limitations make it difficult to ensure a comprehensive dataset of AI-related incidents. 

To address these challenges and enable an exploratory quantitative analysis, the media monitoring tool Cision was employed. This facilitated the extraction of AI-related news coverage in the periods leading up to the respective elections and ensured a robust comparability between varying contexts. 
 

Data Collection

A targeted search using keywords related to artificial intelligence and elections was conducted for each country within a defined timeframe. The number of articles retrieved varied significantly by country, with the United States yielding the highest volume of relevant articles and South Africa the lowest. Country-specific prompts were developed for the searches, maintaining a consistent structure focused on AI, social media platforms, and prominent political parties and candidates. These prompts were tailored to each country's language and localized context to ensure relevance. 

Incident Identification

 The collected articles were manually filtered to identify specific AI incidents. Table X provides an overview of the total number of incidents identified in each country, highlighting that the United States had the highest number of recorded incidents, while South Africa had the fewest. It should be noted that some incidents were widely reported, with multiple articles or headlines referencing the same event. To avoid duplication, each incident was incorporated only once into the analysis. 

Classification of Incidents

The identified incidents were categorized into six threat levels based on the DISARM framework outlined in the European External Action Service (EEAS) report. While the EEAS framework includes five threat categories, an additional category was introduced during the analysis. This sixth category accounted for cases where AI usage had political implications but did not align with the predefined EEAS threat categories. For India, South Africa, and Mexico, local project partners conducted the categorization to ensure contextual accuracy. All other case studies were conducted by the DGAP. 

Limitations and Considerations 

The reliance on newspaper articles for identifying AI incidents inherently introduces a selection bias. Only incidents reported in the media were included, meaning that the dataset likely captures larger, more publicized incidents while overlooking smaller-scale or less-publicized events. This limitation is particularly pronounced for incidents occurring exclusively on social media platforms, which may not have been reported by mainstream media. 

It is essential to emphasize that the resulting list of AI-related incidents is not exhaustive. Instead, this study seeks to move beyond a qualitative focus on selectively known cases by employing an exploratory quantitative approach. While this methodology enables the identification of AI-related incidents on a broader scale, it cannot fully capture all political AI incidents preceding the elections. Nevertheless, the incidents covered by the media are arguably the most impactful and relevant to the public discourse, offering a critical starting point for developing a comparative framework for analyzing AI incidents globally. 

Interactive Map

Explore the Findings 

This interactive map offers a comprehensive overview of the AI/Democracy Initiative’s findings. Discover global trends and country-specific insights to gain a nuanced understanding of the challenges AI brings to electoral contexts worldwide. Navigate using the world map below or dive into detailed country-specific data.  

The Friedrich Naumann Foundation and the German Council on Foreign Relations have collaborated on this project.