Artificial Intelligence and European Elections

刊物介绍

《人工智能资讯周报》探讨人工智能对公共政策、治理和政策建议的影响，探索人工智能对商业、政治和社会的影响，以确定潜在的研究领域，探讨可能的合作研究和机构伙伴关系。本刊着重提供中国人工智能发展动态和对人工智能的思考，同时关注全球范围内人工智能相关研究动态。本刊旨在通过可靠的研究，来帮助企业、研究机构和公民预测和适应技术引领的变化。 和成就。自2017年起，每年的5月30日被设立为“全国科技工作者日”，以鼓励和支持科技工作者的创新和贡献。

Abstract

Technological solutions in European elections demonstrate a significant trend toward internationalization. On one hand, American technology companies such as Smartmatic, NationBuilder, and Blue State Digital provide extensive technical support. On the other, European countries are gradually developing domestic solutions, such as Estonia’s Cybernetica in internet voting systems, Hungary’s Datadat in data-driven campaigning, and France’s LMP, Digitale Box, and Fédéravox for localized political activities. The positive impacts of these technological solutions include the digitization, efficiency, and convenience of election processes. They also create opportunities for broader citizen participation in political processes, particularly by addressing geographical or physical voting limitations through online platforms. However, these technologies also bring significant risks, especially regarding deepfakes, misinformation, and election interference. Notably, most of these companies are driven by technology rather than political expertise, having been established and operated without the input of political specialists. This disconnect between technology and social realities may exacerbate the political and ethical issues arising from the misuse of technological tools.

Electronic Voting Systems

Electronic voting encompasses voting and/or vote counting through electronic means (internet voting). Internet voting, also known as “remote electronic voting” or “online voting,” is usually conducted outside traditional polling stations and before the traditional election day; in other words, it is a geographically and temporally remote voting option, much like postal voting. Recently, electronic voting has been used in Belgium, Bulgaria, Estonia, France, and the Évora region of Portugal. In Bulgaria, voting machines even sparked political controversy. Countries like Italy, Ireland, and the Netherlands have attempted electronic voting but have since discontinued it.

Estonia: i-Voting, Domestic Technology Solution

Estonia has one of the most advanced electronic voting systems in the world. Internet voting—where voters use their own devices to cast votes remotely—was piloted in 2005. The Estonian Parliament passed legislation allowing voters to authenticate themselves using the Smart-ID application and permitting voting via smartphones and tablets. Due to its efficient and convenient nature, internet voting has become extremely popular. Currently, about one-third of votes are cast via the Internet. The current i-voting system was launched during the 2017 local elections, designed by the Estonian company Cybernetica, and has been continuously improved based on new electronic voting policies. To date, no malicious software or security vulnerabilities have been detected during elections in the country.

Bulgaria: Smartmatic, U.S. Technology Solution

Since 2014, Bulgaria has been rolling out electronic voting nationwide, primarily using hardware, software, and services provided by the American election technology company Smartmatic. The hardware, such as the advanced Smartmatic voting machines, includes multiple layers of security, including tamper-proof data collection, end-to-end encryption, and fully auditable logs. Its “end-to-end” encryption technology prevents intrusion and hacking. The machines also generate auditable paper records to ensure the accuracy of digital ballots. Equipped with 17-inch touchscreens, the machines are user-friendly for many disabled individuals. They also include smart card readers and simple printers. The software, known as the Election Management Platform (EMP), is a solution for managing the entire election process. Comprised of multiple modules, it allows election commissions to coordinate election management tasks. In Bulgaria, EMP helps configure voting machines, count, tally, and merge results, and prepare election progress reports. Smartmatic provides project management for the entire election process, recruitment and training of operators, setting up warehouses and service desks, configuring voting equipment and applications, and establishing counting centres.

In 2021, over 9,500 Smartmatic voting machines were deployed in more than 9,000 voting centres across Bulgaria for each round of national elections. Post-election audits confirmed that the digital results were 100% accurate. In the second round of the presidential election held on November 21, 2021, Bulgaria installed 400 machines in its embassies across 21 countries, marking another milestone. Smartmatic, in collaboration with the Bulgarian company Ciela, provided more than 9,500 electronic touch-screen machines for the 2021 July general elections and the November 2021 joint elections and the presidential election, creating the DZZD Electronic Voting Alliance to manage all critical processes for the successful execution of the project.

Digital-Driven Political and Campaign Technology:

Campaigns are defined in the EU’s Study on the Impact of New Technologies on Free and Fair Elections as “organized efforts during an election period where politicians try to persuade people to vote for them by influencing their decision-making process.” In this process, political parties use various channels and media to reach voters as broadly as possible. In countries like the Netherlands, Norway, France, Italy, Spain, and Germany, scholars and journalists have observed that parties in modern campaigns have used targeted advertising. Some Western European parties have recruited American campaign strategists to professionalize their campaigns, particularly in the use of social media and online political micro-targeting. However, this is often less advanced than in the U.S., or at least not as mature across the Atlantic. Despite doubts about the effectiveness of micro-targeting, research by Almog Simchon and others shows that political micro-targeting in Europe, while limited in its persuasive effect, still holds potential. Companies like CrosbyTextor, Blue State Digital, and Harris Media have worked with multiple European parties, frequently (but not always) collaborating with parties that share similar ideological agendas.

NationBuilder

NationBuilder is widely used globally across different political spectra. The company’s listed clients include Austria’s Junge Linke (Youth Left), the Alliance of Liberals and Democrats for Europe (ALDE), and Belgium’s Christian Democratic Party (CD&V). The UK’s Information Commissioner’s Office (ICO) revealed that during the 2017 UK general election, up to 200 UK campaign groups used NationBuilder.

NationBuilder offers a suite of tools for tracking supporter behavior, campaigning, and outreach. It can be used for both external voter mobilization and organizing supporters. Its two main features are its database, known as Nation, and its user analysis tool, Political/Social Capital. The database serves as a centralized repository for data from various sources and organizational units. Clients can import supporter data from party internal records, public records, marketing, and online interactions. NationBuilder does not (or at least by default does not) match social media accounts with the email addresses of supporters in the database. Its user analysis tool categorizes and scores activities both within and outside the platform. More than 80 different actions influence the Political/Social Capital metric in the database. Political/Social Capital can be understood as a measure of engagement, influenced by positive interactions like donations, liking Facebook posts, or agreeing to become fundraisers, and negatively affected by actions like violating internal rules.

NationBuilder also offers optional add-ons and customization options. Through integrations with Statara Solutions, Accurate Append, and NamSor, clients can add supplementary data to supporter files. Statara Solutions also provides custom predictive models and analytics to help organizations expand recruitment, fundraising, and outreach. Clients can also integrate with data suppliers via intermediaries, connecting ad workflows to Amazon and Google ads using intermediaries like Pipedream and Zapier.

Although campaigners globally use NationBuilder to promote legitimate causes, it may also be misused for harmful purposes. Research on the global installation of NationBuilder in 2017 identified three suspicious uses: (1) as a tool for inciting hatred or mobilizing against cultural or racial identities; (2) for deceptive advertising or hidden media profiles; (3) for fundraising in corporate journalism.

France

Local digital campaign companies in France include LMP, founded in October 2012, Digitale Box, established in August 2013, and Fédéravox, created in June 2014. Currently, fewer than a dozen companies are involved in the French market, with American company NationBuilder having the most developed and structured internal organization, boasting over 30 business segments related to data, digital, and commercial and territorial development.

These companies offer two main services to clients: data software services and political consulting. The first is data-based software services (e.g., Ecanvasser, DigitalBox, NationBuilder, Liegey Muller Pons, Spallian, and Fédéravox), which require computing and/or statistical skills. The software services vary widely but are mainly focused on targeting features of varying precision. Targeting can be name-based (e.g., in CRM systems such as Ecanvasser, DigitalBox, NationBuilder, Liegey Muller Pons, and Fédéravox). Depending on the complexity of the software, especially the richness of available data in the database, clients can cross-reference different types of information (e.g., GPS coordinates, customer preferences, phone numbers, and socio-demographic data) to optimize campaigns according to their strategy. Typically, the software also offers data visualization features, including maps. When data is insufficiently precise, users can infer data indirectly through predictive models (e.g., P-A. Veillon, Liegey Muller Pons, Fédéravox). The second service is political consulting based on digital tools, including software training (e.g., E. Zuckerman, LMP, Fédéravox). Additionally, some clients offer on-site training (for field workers) to implement campaigns, such as canvassing (e.g., LMP).

l Diversified service providers: Liegey Muller Pons and Fédéravox offer comprehensive services, covering CRM, cartography, targeting, and software training, meeting the varied needs of clients. This integration also enables them to serve a broader client base and gain more value from data integration.

l Companies focused on specific functions: E. Zuckerman and PA Veillon specialize in providing specific services such as targeting and consulting, showing their professional focus in niche markets. Their primary clientele consists of political organizations, indicating their vertical market positioning.

l Technology-driven companies: NationBuilder and Digitale Box not only provide conventional CRM and targeting services but also expand into software engineering and platform strategy development, demonstrating their strong capabilities in technological innovation. These companies are likely better suited for clients who need deep technical support, especially those aiming for data integration and optimization through platform solutions.

The services and data usage types of participants in France’s digital campaigning are detailed as follows:

Nation Builder

Services: Customer Relationship Management (CRM), Targeting, Software Training, Software Engineering Services, and Platform Strategy.

Data usage: Field Data, Socio-economic Data.

E Zuckerman

Services: Cartography, Targeting, Predictive Models, and Consulting (exclusively for political clients).

Data type: Socio-economic Data.

PA Veillon

Services: Targeting, Consulting, and Software Engineering Services.

Ecanvasser

Services: Customer Relationship Management (CRM), Targeting, Software Training, and Platform Strategy.

Data usage: Field Data.

Digitale Box

Services: Customer Relationship Management (CRM), Targeting, Software Training, and Platform Strategy.

Data types: Socio-economic Data, Electoral Lists.

Liegey Muller Pons

Services: Customer Relationship Management (CRM), Targeting, Cartography, and Software Training.

Data types: Socio-economic Data, Electoral Lists.

Spallian

Services: Targeting.

Data type: Field Data.

Fédéravox

Services: Customer Relationship Management (CRM), Targeting, Cartography, and Software Training.

Data types: Field Data, Socio-economic Data, Electoral Lists.

Hungary

During Hungary’s 2022 elections, data collected through public services (such as tax benefits, COVID-19 vaccinations, and professional associations) was utilized in the ruling Fidesz party’s campaign. The opposition also adopted data-driven campaigning during the 2019 elections, particularly through collaboration with the digital campaigning company Datadat. This company, founded by former government officials, includes notable figures such as former Prime Minister Gordon Bajnai, former Intelligence Minister Ádám Ficsor, campaign strategist Viktor Szigetvári, and sociologist Tibor Dessewffy.

Datadat has worked with various opposition parties and municipal candidates in Hungary, as well as provided campaign support in other countries such as Romania and Italy. According to the independent news outlet Átlátszó, during the 2019 Hungarian municipal elections, Datadat was responsible for targeted Facebook ads for the newly established news website EzaLényeg. The site featured carefully curated and targeted information designed to resonate with specific voter groups. The report also noted that several Democratic Party candidates listed Datadat as their data processor. The online news portal 24.hu revealed that Datadat’s chatbot enabled campaign teams using it to record user feedback, providing insights into voters’ views on specific issues and their importance. This allowed highly precise targeting of political messages and helped parties build online communities.

The Netherlands

In the 2017 Dutch elections, the Green Party hired the U.S.-based digital strategy company Blue State Digital, founded by former digital strategists for President Barack Obama. The campaign strategy was designed by Green Party leader Jesse Klaver and Wijnand Duyvendak. Their campaign team modeled their approach after U.S. Democrats and utilized software for organizing and communicating with activists.

Applications of Artificial Intelligence (AI):

AI can be used to enhance election security and management, combat misinformation, support resource-strapped campaigns, improve political advertising, moderate content, debunk fake news, and provide voters with election-related advice and education. On the other hand, AI can be exploited to create harmful content at scale and at minimal cost. Campaigners might attempt to excessively discredit opponents with deepfake content, polarize and radicalize voters through misinformation, deter voter participation using robocalls and cloned voices (“rumor bombing”), or avoid politically damaging scandals by framing truthful information as false (“liar’s dividend”).

Since late 2022, generative AI systems have surged, enabling users to create increasingly realistic synthetic images, audio, video, and text from simple prompts. Large language models (e.g., GPT-4) and image generators (e.g., Midjourney AI) offer promising tools for underfunded political campaigns and election management. For example, chatbots can enhance connections between political candidates and voters by translating outputs into multiple languages. However, the same innovations lower barriers for technically unsophisticated but malicious actors, who previously lacked access to such tools.

Generative AI also accelerates campaigns’ speed, scale, authenticity, and personalization, influencing audiences in various ways, including:

l Personal Attacks: Highly realistic AI-generated content containing false allegations designed to damage political candidates.

l Deceptive Political Ads: AI-generated content misrepresenting candidates’ appearances, endorsements, or events to deceive voters.

l Voter Targeting: Automated bot accounts and targeted voter interactions aimed at shaping public opinion.

United Kingdom

According to research by the Centre for Emerging Technology and Security (CETaS), 16 confirmed cases of AI-generated manipulation or deepfakes have been identified in 112 national elections in the UK since January 2023 or scheduled for 2024. These findings suggest that generative AI content or deepfakes have not significantly impacted UK election outcomes.

France

In the lead-up to the June 2024 European Parliament elections, parties affiliated with the far-right Identity and Democracy (ID) alliance integrated generative AI into their campaigns. Research by DFRLab revealed that France’s National Rally and Italy’s League (both ID member parties) used generative AI to amplify their political messages. Since at least the fall of 2023, ID parties have used tools like Midjourney and Stable Diffusion in dozens of instances. The AI-generated synthetic images, promoted via social media, depicted scenarios such as alleged immigrant “invasions” of the EU, tractors participating in farmer protests, or unflattering portrayals of EU politicians. These posts violated the voluntary Code of Conduct signed by ID in April 2024, which included commitments to label AI-generated content and adhere to transparency guidelines.

Research on the 2024 French elections found limited use of generative AI tools in campaigns. Most official party or alliance channels showed little evidence of AI-generated images. However, AI Forensics identified systematic use of generative AI images in the National Rally’s campaigns, along with evidence of usage by Reconquête, Les Patriotes, L’Alliance Rurale, and Union des Démocrates et Indépendants. Across all platforms and official party channels, AI Forensics discovered 51 posts containing AI-generated images, 25 of which were unique.

The use of AI-generated images in French political campaigns contradicts commitments established in the Code of Conduct for the 2024 European Parliament elections. None of these images were labeled as AI-generated by user accounts (i.e., political parties) or platforms. Transparency and integrity in political campaigns rely on platforms clearly labeling AI-generated content. This lack of transparency violates the Digital Services Act (DSA) guidelines on election integrity. These guidelines mandate platforms capable of generating deceptive AI content to employ robust detection methods. They also require clear labeling of synthetic or manipulated images that resemble real individuals, places, or entities. The failure to label and detect AI-generated content undermines efforts to combat misinformation, protect democratic processes from manipulation, and foster informed public engagement through critical participation and media literacy.

On the one hand, electronic voting systems can enhance efficiency and convenience. Electronic voting processes are swift and effective; for instance, they can reduce the labor intensity of vote counting and potentially accelerate the process. In the context of remote voting, electronic systems are generally more universally applicable and efficient than other geographically and temporally restricted forms of voting, such as postal voting. This also reduces reliance on increasingly limited postal services. Moreover, digitization is a prevailing societal trend that seems inevitable even in elections. Estonia’s internet voting system, while not universally supported (criticized domestically by parties like the Centre Party and internationally by organizations such as the OSCE), appears to be part of the country’s broader transition to digital citizenship. This transition encompasses services ranging from digital identities to administrative affairs, extending far beyond the electoral realm.

AI technology can also serve as a catalyst for democracy by increasing political participation and transforming campaign strategies. AI can help citizens better understand policy issues or familiarize themselves with politicians’ positions. For example, chatbots based on political recommendation systems can answer citizens’ questions about candidates’ platforms. Additionally, specialized AI tools can provide real-time updates on policies of interest, enabling citizens to express their views more effectively when engaging with governments and politicians. AI can also improve civic debates by managing large-scale political discussions in chatrooms, such as summarizing participants’ opinions, mediating disputes by identifying tension points, steering conversations away from insults and attacks, and even serving as a consensus builder. Furthermore, the methods of political campaigning might undergo significant changes. Politicians will soon be able to use AI to respond instantly to campaign dynamics. Frequent use of sentiment analysis on social media will enable them to better understand the issues that concern and interest citizens. AI could then be used to generate personalized emails or text messages via chatbots for specific audiences. However, this also means politicians can very precisely target specific groups, including swing voters.

On the other hand, AI and data-driven technologies present well-known challenges, such as technical misuse, deepfakes, and misinformation. The OECD has emphasized that combining AI language models with misinformation could lead to large-scale deception and erode public trust in democratic institutions. AI systems themselves are capable of generating false information. One study found that Google’s Bard AI tool generated persuasive misinformation in 78 out of 100 tested narratives. Similarly, the newer version of ChatGPT (GPT-4) is more prone to creating misinformation than its predecessor (GPT-3.5) and does so with greater persuasiveness.

It is also noteworthy that a lack of political expertise characterizes the current digital campaign industry in Europe and the U.S. “In my company, I have only programmers and developers, not a single political scientist or campaign director,” explained Vincent Moncenis, founder of Digitale Box. This absence highlights the primacy of technology over politics in the digital campaigning domain. A similar logic can be observed at NationBuilder, which chose to position itself as a “technology company” focused on technical solutions. In the absence of political experts, the company built an ecosystem around its platform, enabling clients to easily hire such experts for their campaigns. This suggests that the rapid rise of many digital campaigning technology companies is not based on their political attributes.

文章推送方式:

微信：关注“海国图智研究院”公众号

网站：收藏我们的网站 www.intellisia.org

邮件：扫描下方二维码登记邮箱信息，每日接收新鲜、深度的时政洞察

海国图智研究院(Intellisia Institute)是中国第一批独立的新型社会智库之一。海国图智专注于国际问题研究，并主要聚焦中美关系、中国外交、风险预测、新科技与国际关系等议题，致力于通过书目与报告的出版、学术与社会活动的组织、研究项目的承接和开展等形式为政府、企业、媒体、学界社会公众提供知识资源，以帮助其更好地“开眼看世界”，了解中国与世界的关系，为其对外事务提供战略见解和政策解决方案。