German debate on AI governance: Insights from network analyses

*The original of this text was published in German on 1 February.*

European institutions have been working on a regulation for artificial intelligence (AI) called the AI Act for several years. After the Council of the European Union reached a common position in December 2022, the AI Act is still facing intense debates in the European Parliament. Despite multiple delays due to unresolved questions, it is expected that the regulation will be passed later this year. This would establish the first regulatory framework for algorithmic systems, commonly referred to as “artificial intelligence,” in all 27 EU member states, including Germany. But does this mark the beginning and end of political discussions surrounding AI regulation in Germany? And what insights can we gain from network analyses?

AI made in Germany – They were working on this, right?

Germany’s AI policy journey began in 2017 with the introduction of the national strategy “AI Made in Germany,” operating independently from the European level. Since then, thirty-three policy projects have been initiated, positioning Germany in the upper midfield among countries actively engaged in AI policy, trailing behind the United States and the United Kingdom, according to OECD.AI. However, there has been a decline in such activities in recent years. Even after the 2021 federal election, we see no significant shift in the German government’s fundamental stance on AI.

Compared to other policy areas, German AI policy is still in its early stages. Policymakers face challenges when addressing complex issues that have received limited political attention so far. This is especially true for technological developments in artificial intelligence, which are accompanied by substantial uncertainty regarding key issues and possible effects. A diversity of perspectives and information is thus crucial, particularly in the initial stages, to a thorough analysis of the various aspects of the issue that allows policymakers to formulate potential policy solutions. Finding the right balance between a diversity of perspectives and information and making clear, coherent policy decisions is the primary challenge for government action in dealing with complex issues. Consequently, the involvement, or lack thereof, of organizations in the ongoing regulatory and implementation process becomes crucial. Which organizations played a role in the early stages of the AI political discourse in Germany? What aspects of the topic were prioritized?

Insights from discourse network analyses

To gain insights into the discourse surrounding AI in Germany, an analysis was conducted on contributions made during the 2018 government consultation on the German AI strategy. The consultation involved the participation of 91 organizations from academia, industry, and civil society. A total of 818 comments were submitted across twelve predefined topic areas, with many organizations providing comments on all of them. For analysis purposes, the comments submitted by each organization were examined to determine the topic areas they emphasized in relation to the AI strategy. The resulting discourse network diagram depicts the aggregated results: organization types are represented by gray squares, while the topic areas mentioned are denoted by blue circles. The strength of the connection between an organization type and a topic area indicates the frequency with which organizations from this category mentioned the respective topic area.

As expected, the topic of technology garnered significant attention from all participating organizations. However, a more intriguing aspect is the emergence of a cluster of companies, business interest groups and scientific organizations in the center of the discourse network. These entities not only emphasize technology but also highlight the topic of domestic commerce. Additionally, companies and business interest groups place emphasis on education, particularly in training future AI experts. All three organizational groups also underscore the topic area of government operations, which includes the governance of AI among its focus areas.

Companies and business interest groups make up half of the participating organizations, totaling 45. Many of these companies, spanning various sizes, come from Germany’s digital sector. Alongside digital sector representative, business interest groups encompass organizations from other crucial industrial sectors in the country. When scientific organizations are included, this central group in the discourse network represents over two-thirds of all government consultation participants. This group includes numerous universities, expert groups from the Digital Summit, and other important players in German research, such as the Fraunhofer Gesellschaft and the German Aerospace Centre.

In contrast, the category of public interest groups is much smaller, consisting of only seven actors, and appears to have less involvement in discussion concerning the economic aspects of AI. This category includes organizations like the German Federation of Trade Unions (DGB), the United Services Union (ver.di), the German Federation of Consumer Organizations, and the German Environmental Aid Association (Deutsche Umwelthilfe). Notably, certain topics receive less emphasis overall. These include areas directly impacted by AI’s potential application, such as labor and civil rights, as well as domains like health and social welfare, which represent potential sectors for the implementation of AI applications.

In summary, the participation of companies and business interest groups in the 2018 government consultation was notably robust, whereas civil society voices were comparatively underrepresented. In terms of content, discussions on artificial intelligence at this early stage of the political debate focused primarily on technical aspects, with a stronger emphasis on economic and industrial policy. Social and environmental concerns received less attention. These findings align with an additional analysis of early media and parliamentary discourse on AI in Germany from 2017 to 2019, both preceding and following the government consultation.

Overall, the early political debate on AI in Germany appears fragmented, with few organizations bridging AI with other topics, such as potential application sectors, during this phase. Moreover, civil society voices were initially less prominent in the public policy discourse.

Implications for AI governance

The inclusion of diverse perspectives raises important considerations for AI governance. The availability of diverse information is crucial for policymakers to make informed decisions regarding AI policy. Insufficient diversity in information can create problems, such as overlooking critical aspects of the issue at hand. Already, we have witnessed specific AI applications sparking political controversies and even necessitating their discontinuation. A notable example is the Dutch case of “SyRi” in 2020, which gained global attention. SyRi was a tool designed to aid authorities in detecting fraud in social benefits, allowances, and taxes. However, the indicators used in the tool exhibited discriminatory biases, leading to false accusations of fraud against citizens eligible for child benefits. Such controversies erode public trust, potentially impeding future AI applications and hindering the acceptance of technological innovations.

Involving civil society for more diversity of perspectives

The journey toward effective AI regulation in Germany doesn’t end with the implementation of the AI Act. Many stakeholders will continue to grapple with supplementary initiatives, such as revising the Product Liability Directive, establishing an AI liability directive, addressing AI standardization, and implementing the AI Act itself. Adaptations are also necessary at the national level. Recently, 19 civil society organizations highlighted the need to amend the General Equal Treatment Act (AGG) in an open letter, as its current form fails to protect against algorithm-driven discrimination. The range of AI-related issues requiring decisions is vast and varied. Consequently, the governance of AI should incorporate diverse perspectives, including those from civil society and sectors already impacted or soon to be impacted by AI applications. While this may initially appear complex, it proves beneficial in the long run. Early identification and political negotiation of important and controversial matters are possible through diverse exchanges. Conversely, a lack of communication and coordination resulting from misunderstandings or siloed thinking can negatively impact the implementation of AI innovations. Overall, the inclusion of diverse perspectives, particularly those from civil society contributes significantly to formulating and implementing effective policy solutions for AI. At present, it is uncertain which direction the German government will take in shaping AI governance within the European and international context. Initially, the focus was on a harmonized European approach through the AI Act. However, even within the coalition, differences persist. For instance, the SPD parliamentary group published a position paper in the Bundestag only after the Council of the European Union and all 27 EU member states reached a consensus in December 2022. They called for subsequent improvements, as reported by Tagesspiegel Background. Given the rapid technological advancements in this field, it is crucial that AI governance remains a prominent item on their agenda, seizing important opportunities at the right time.

*The analysis presented in this paper is based on a joint research paper by Nicole Lemke with Professors Philipp Trein (University of Lausanne) and Frédéric Varone (University of Geneva).

This text is licensed under a Creative Commons Attribution 4.0 International License