EU-cooperation in the Development and Deployment of Military Artificial
Intelligence.
Rachel Splinters s1622595 Master Thesis
Leiden University: Faculty of Governance and Global Affairs Supervisor: Dr. mr. E. E. A. Dijxhoorn
Second Reader: Dr. J. Shires Word count: 18.077
2 List of Abbreviations
AI Artificial Intelligence
DFKI German Research Center for Artificial Intelligence LAWS Lethal Autonomous Weapons System
AWS Autonomous Weapons System
CCW The Convention on Certain Conventional Weapons CDU Christian Democratic Union
CSDP Common Security and Defence Policy CSU Christian Social Union
ECAP European Capabilities Action Plan EDA European Defence Agency
ESDP European Security and Defence Policy
EU European Union
FCAS Airbus and Dassault Future Combat Air System HHG Helsinki Headline Goal
MUSAS Maritime Unmanned Anti-Submarine System NATO North Atlantic Treaty Organization
PESCO Permanent Structured Cooperation TEU Treaty on the European Union UAS Unmanned Aerial System
UK United Kingdom
UNSC United Nations Security Council
3 Table of Contents
1. Introduction 4
2. Analytical framework 9
2.1 EU defence cooperation 9
2.1.1 Common Security and Defence Policy 9
2.1.2 Permanent Structured Cooperation 10
2.1.3 EU-NATO relationship 11 2.1.4 Franco-German relationship 12 2.2 Artificial Intelligence 18 2.2.1 General AI 18 2.2.2 Military AI 20 2.2.3 AI in Europe 22 3. Methodology 27 3.1 Operationalization 27 3.2 Data collection 28 3.3 Limitations 29 4. Analysis 31
4.1 Priorities of both countries 31
4.1.1 Germany 31 4.1.2 France 33 4.1.3 Comparison 35 4.2 Military AI capabilities 36 4.2.1 Germany 37 4.2.2 France 39 4.2.3 Comparison 40 4.3 Cooperation 41 4.3.1 Germany 42 4.3.2 France 42 4.3.3 Comparison 44 5. Conclusion 46 References 50
4 1. Introduction
On the 19th of February 2020, the European Commission unveiled its plans to regulate all domains of Artificial Intelligence (AI) to increase efforts to keep up with the United States and China (Amiel, 2020; European Commission, 2019). These recent plans address how the EU can use and develop the full capabilities of AI, while respecting individual rights and freedoms. The Commission's focus is mainly on developing AI for economic and public administration purposes. It is striking that the development of AI for military and defence purposes is not included in these plans since states like the US and China already develop AI for military purposes. During the 2019 Finnish presidency of the EU, however, a Food for Thought paper on military AI was drawn up, which shows the importance of developing AI and the importance of military cooperation between Member States and with other countries or organizations like NATO for the development of military AI capabilities (Hill, 2020; Food for Thought Paper, 2019). Recently, the European Defence Agency (EDA) (2020) also launched a preliminary blueprint to promote and coordinate AI innovation across its Member States, which has been unfolding in phases.
Because of the unique advantages AI offers and its many interfaces with cyber, economic, environmental, infrastructure and military security concerns it is an important dimension of European security. According to Federica Mogherini, vice president of the European Commission, the EU recognizes that some areas of AI pose new security challenges (Banks, 2018). It seems contradictory that the EU believes that the development or deployment of AI in the world is a matter of security, but at the same time does not proceed in developing military AI capabilities and policies. Technological developments lead to both new risks and opportunities for Europe. AI increases the effectiveness of logistics, planning and transportation in the military, but can eventually also be used to select and kill a target even without human control. In the current security environment, Member States can keep Europe safe through developing a full spectrum of military capabilities. This can be established by more defence cooperation because no European state can deal with these challenges on its own (Bakker, Drent, Landman & Zandee, 2016; European Defence Agency. 2018).
EU defence cooperation is not a new topic, but the EU Treaty (Lisbon Treaty – 2009) made it possible to raise EU-cooperation to a higher level. The EU Security Strategy in 2016, again, made defence cooperation a top priority with an increasing focus on the development of military capabilities that can improve the capacity to conduct missions and operations (Blockmans,
5 2018; Ilinca, 2020). Despite multiple initiatives, there are challenges related to defence capability development in the EU (Fiott, 2018). EU defence cooperation depends on national defence policies, which can ensure greater convergence or divergence of national policies that will determine the future success or failure of European defence cooperation. In order to avoid fragmentation of European collaborative efforts and to coordinate disparate national policies, the EU is encouraging greater European defence cooperation (Fiott, 2018; Keohane, 2018). The EU does this mainly through the Permanent Structured Cooperation (PESCO) framework.
On 12 November 2019, EU defence ministers approved a third wave of 12 new proposals under the PESCO framework, resulting in a higher number of joint military projects (Brzozowski, 2019). There is, however, still little EU-cooperation in the development and deployment of military AI. Only one of the new PESCO projects, the Maritime Unmanned Anti-Submarine System (MUSAS) project, aims to develop an advanced command, control and communications service for anti-submarine warfare by taking advantage of artificial intelligence. Therefore, this thesis is examining the main hurdles the EU-cooperation faces for the development and deployment of military AI capabilities. This has led to the formulation of the following research question: What are the hurdles for the EU-cooperation in the development and deployment of military AI capabilities? This thesis focuses on the hurdles France and Germany create for the development and deployment of military AI capabilities in the context of the EU the same way as with the development and deployment of other military capabilities. Since 2014, several internal and external factors, such as the United Kingdom’s vote to leave the EU, the election of the US President Donald Trump and the challenging security environment in the EU caused by the situation in the Middle East and North Africa, have deepened the development of EU defence cooperation between France and Germany (Blockmans, 2018; Brattberg & Valásek, 2019; Daehnhardt, 2018; Kempin & Kunz, 2017; Keohane, 2018). After the UK’s withdrawal from the EU, France and Germany are the backbone of European defence as they represent about 50 percent of the military and industrial capabilities within the EU and about 40 percent in wider Europe (Daehnhardt, 2018; Major & Mölling, 2018). However, cooperation in the development deployment of military capabilities between the two countries, such as the development of battle tanks, was most of the time not successful because of major political and cultural differences. By examining the different positions of France and Germany on military AI in the context of the EU, it becomes clear whether there are still differences that prevent both countries from cooperation in the field of military AI the same way as in in the development and deployment of other military capabilities.
6 This thesis only focuses on the hurdles for EU-cooperation in the development and deployment of military AI capabilities. The EU, however, still has to deal with issues that continue to exist when the hurdles for the EU defence cooperation on the development and deployment of military AI capabilities are removed. Due to this, my conclusions are formed from a certain angle. The EU is constantly confronted with challenges, such as the lack of political and financial commitments of Member States, the lack of a common strategic culture due to diverging threat perceptions, and the lack of transparency and accountability in security and defence matters. One of the biggest challenges for EU-cooperation is the defence budget (Csernatoni, 2020; Kempin & Kunz, 2017). Defence spending is both collectively and individually inefficient which contributes to a lack of interoperability in the EU. After many Member States had removed a significant amount of defence funding, the focus is mainly on the national defence instead of the European defence industry. Brexit also causes a decrease in the EU defence budget as the UK was one of the largest contributors to the EU budget. Recently, many Member States were increasing the defence budgets in order to invest in the development and deployment of new military technology innovations. However, theCovid-19 pandemic is slowing down this process (Csernatoni, 2020). Due to the pandemic, countries have to alter their spendings. Recent governmental tendencies to spending in times where the economic situation is not at its best, are a way of setting new priorities.
The second challenge that remains a problem, are the differences between Member States when it comes to the national defence industries. The Member States with strong national defence industries and the means to co-finance projects are also expected to be the main beneficiaries of the EU defence industrial integration. For this reason, smaller Member States are less likely to fund defence projects since larger States are most likely in charge of the projects (Csernatoni, 2020). Thirdly, the goal of the EU is to become a sovereign player in security and defence, however many EU Member States want to maintain some degree of freedom of action and sovereignty, which limits EU defence cooperation (Csernatoni, 2020). For many Member States, NATO also remains the organization responsible for territorial defence and not the EU. So, it is difficult to make strategic autonomy as a narrative for the European political project. Next to those existing challenges, there are other factors that make EU-cooperation in the development and deployment of military AI difficult, such as the General Data Protection Regulation (GDPR). The regulation focuses on consumer personal data privacy and protection. GDPR's restrictions on personal data ensure that the development of AI is limited. This may
7 lead to the EU falling behind in its AI efforts in all domains because of more regulations (Baladi, 2019; Humerick, 2018). For future research it would be interesting to take these above-mentioned factors into account, and to set out the process of EU defence cooperation in the development and deployment of military AI in greater detail.
This thesis has social relevance, because according to Csernatoni (2019) there is a sense of urgency that more investments must be made in the research and development of disruptive AI defence applications (Csernatoni, 2019). China, Russia, and the US are increasingly investing in AI technology and are already developing military AI capabilities. Even smaller countries are heavily involved in the development and use of AI technology in military capabilities such as Israel and South-Korea (Slijper, Beck & Kayser, 2019). Whilst Europe is increasingly lagging behind on the matter, this increasing amount of threat can result in a competition for global leadership in the field of military AI (Leonard & Röttgen, 2018).
Currently, the impact of AI on national security and military affairs is receiving more attention in AI research in both the technical and political realm. Research has shown that new weapon technologies can increase the defence and military capabilities of a state (Burton & Soare, 2019; Dahab, 2018; Garcia, 2016; Sökmen Alaca, 2019). New weapon technologies have the potential to change the nature and character of warfare as well as law enforcement because it can increase the speed and quality of military capabilities. Studies about the military applications of AI are, however, mainly focusing on the US, China or Russia. Europe tends to be overlooked. This means that there is little information about European thinking on military AI. However, research on military AI in the context of the EU is important, as the EU has made a lot of progress in the development of an AI strategy in 2019. Next to that, many Member States are developing or have implemented national policies and programs about the use of AI and several European companies are already developing AI-enabled military systems (Franke, 2019; Slijper, Beck & Kayser, 2019).
The research is relevant in the context of crisis and security management as technological innovations can influence the global distribution of power and international conflict (Garcia, 2016; 2018). As long as the EU postpones its policies about the development and deployment of military AI applications while focusing more on providing ethical guidelines, it will lag behind while the others continue to develop military AI. In this time, other powerful states are already building a deadly AI-enabled weapon (Csernatoni, 2019). With technological
8 innovations like military AI, effective governance is needed to maximize the benefits and mitigate the risks. This thesis hopes to contribute to the academic literature on EU defence cooperation, in particular by providing a more in-depth look on the EU-cooperation on military AI. Besides its contribution to the academic debate, this research also contributes to the explanation of the reason why military AI is absent from the EU’s recent AI strategy by comparing the positions on military AI of the EU’s biggest players in defence. By doing so, this thesis hopes to broaden the knowledge and insights concerning military AI. Furthermore, because not much research has been done on military AI, this research is an extension of existing literature. This thesis can be relevant for the EU to gain insight into the effectiveness of policy-making of certain security issues.
In order to answer the question what the hurdles for EU-cooperation in the development and deployment of military AI capabilities are, this thesis will first set out how EU defence cooperation developed into what it is today, based on a literature review. In addition, the Franco-German relationship and its effectiveness are discussed in more detail. Second, based on the literature review, it is explained what the complex term AI means, and how this manifests itself in a military context and the development of military AI in the context of the EU. Based on the literature on EU defence cooperation and military AI, an analytical framework has been formed. Thirdly, the methodology is explained. For this thesis a comparative document analysis is done. By comparing documents containing information of both the French and German position on military AI it is examined whether their positions differ on the matter and how their position relates to previous military cooperation issues. These differences could challenge EU-cooperation in the field of military AI, as France and Germany are the engine for European integration. Fourth, the results of the analysis are set out, which explain the different views of France and Germany on the development and deployment of military AI capabilities. This leads to the conclusion that the hurdles for EU-cooperation in the development and deployment of military AI capabilities arise because of the different strategic cultures of France and Germany. These remaining differences in strategic cultures make EU defence cooperation in the development and deployment of military AI capabilities limited.
9 2. Analytical Framework
In order to answer the question what the hurdles for the EU-cooperation in the development and deployment of military AI capabilities are, this section will discuss the different concepts that are central in this research. First, the emergence and development of EU defence cooperation is discussed. An analytical framework is being set out by formulating the different factors that contributed to the successes and failures of EU-cooperation for the development of other military capabilities in both France and Germany. Second, the concept of AI in general and military AI in particular is explained including a review of the literature on military AI. In addition, AI in Europe is discussed to give an idea of where the EU currently stands in the development and deployment of AI.
2.1 EU defence cooperation
2.1.1 Common Security and Defence Policy
In 1999, the beginning of defence cooperation in the EU arose by the creation of a European Security and Defence Policy (ESDP), following an initiative of the UK and France, resulting in the 1998 St. Malo Declaration (Greco, Pirozzi & Silvestri, 2010; Keukeleire & Delreux, 2014). The two countries intended that the EU should create the capacity for autonomous action through credible forces and the means to use them (Greco, Pirozzi & Silvestri, 2010).
During the Cold War, cooperation among European States in the development of military capabilities has remained more limited than it is today (European Parliament, 2020). Indeed, there was a national focus on military capabilities, as the Member States were allowed to develop their own capabilities. This changed after the Cold War as it was recognized that the EU itself developed and used few military capabilities (European Parliament, 2020). There is no generally accepted definition for the term military capability. According to NATO, it has several meanings. On the one hand, it comes down to the scale and quality of military power compared to other superpowers or external threats. On the other hand, it also involves the development and deployment of certain aspects of the military, such as organization, training and material (European Parliament, 2020). Military capabilities refer to the aims and means of military action which are needed for the implementation of military strategies (Greco, Pirozzi & Silvestri, 2010).
A milestone for the deepening of EU-cooperation in the development of military capabilities was the Helsinki Summit in December 1999. During the Helsinki Summit, the Member States
10 agreed upon the establishment of EU military capabilities with a collective capability goal, called the Helsinki Headline Goal (HHG); (Helsinki European Council, 2012). This collective capability goal led to the launch of the European Capabilities Action Plan (ECAP), which explained how the EU can improve the effectiveness and efficiency of European cooperation in the field of defence (ECAP, 2018).
Europe increasingly took the initiative in defence cooperation by conducting various operations in the Western Balkans and Africa, developing the first European Security Strategy (ESS) in 2003 and establishing the EDA which supports the efforts made by Member States in developing military capabilities (Ilinca, 2020). However, the EU defence cooperation really got a boost after the ESDP was formalized and relabelled as the Common Security and Defence Policy (CSDP) in the Lisbon Treaty. The CSDP is about the development of new instruments, while not only focusing on the performance of military crisis management operations, but also on civilian crisis management operations such as natural and humanitarian disasters (Juvan & Prebilic, 2012). Since then, the EU has drawn up many plans and strategies for European defence and security.
2.1.2 Permanent Structured Cooperation
The Lisbon Treaty already included a possibility for permanent structured cooperation, but EU defence cooperation has only recently seen a major upswing because several international crises have put military capability development back on the political agenda. New initiatives such as the Capabilities Development Priorities (CDP), the Coordinated Annual Review on Defence (CARD), the European Defence Fund (EDF) and the Permanent Structured Cooperation (PESCO) are at the centre of this discussion (Drent & Zandee, 2018). The CDP clarifies which capability priorities member states should focus on, where PESCO provides options for how these capabilities can be developed in a collaborative manner (European Defence Agency, 2018). CARD provided reviews of Member States' defence plans, the implementation of the CDP priorities and the development of European cooperation (Ilinca, 2020). EDF is the first European funding for military research and procurement (Marrone, 2017).
The Council has activated PESCO in 2017 with 25 out of 27 Member States participating. When accepting the PESCO frameworks, investment, planning, development and operating of defence capabilities are made possible (Billon-Galland & Efstathiou, 2019). In Articles 42 (6) and 46 of the Lisbon Treaty, PESCO’s aim is described to be an effective tool in order to improve
11 integrated European projects in the field of defence through deepened cooperation between a smaller group of countries (Billon-Galland & Efstathiou, 2019; De France, Major & Sartori, 2017; Ilinca, 2020; Kempin & Kunz, 2017; Keohane, 2018). Closer defence cooperation will make the EU's defence sector more independent of the US. PESCO is a legally binding framework that requires states to meet their commitments, but the challenges of a lack of interest and political will does not disappear. Often, cooperation partners still follow national interests rather than common interests in the context of the EU (De France, Major & Sartori, 2017). Cooperation between Member States is often based on trust built up by previous collaborations, which can create challenges in defence cooperation between states that have had little close cooperation (European Parliament, 2020).
On 12 November 2019, EU defence ministers decided to launch 13 new proposals for joint military projects under the PESCO framework (Brzozowski, 2019). The projects mainly focus on training projects in the field of cybersecurity and space such as Cyber Rapid Response Teams, EU Training Mission Competence Centre and the creation of a European Military Space Surveillance Awareness Network (Brzozowski, 2019; Ilinca, 2020). However, there is a lot of scepticism for the success of PESCO projects, as Member States are struggling to achieve the military level of ambition under the CSDP which will result in capability shortfalls (Barrie et al., 2018; Billon-Galland & Efsathiou, 2019). Based on the difference in security threats between the Member States, different capability priorities are also emerging (European Parliament, 2020).
2.1.3 EU-NATO relationship
EU defence cooperation cannot be dissociated from the transatlantic security link with the US (Daehnhardt, 2018). It is therefore important to clarify the relationship between the EU and NATO, who ensures which security? Defence cooperation among EU members mainly took place outside the EU framework, mainly through NATO. However, EU defence cooperation is about the merging of national defence policies, rather than about the relationship with NATO (Keohane, 2018). Where the EU is not a military alliance, NATO is an alliance of thirty states with a collective defence as one of its core tasks (Hill, 2020). The EU has long been dependent on the US-dominated NATO when it comes to the development of military capabilities, because NATO is the biggest international player in international security and peace. Since the EU has to deal with some security challenges on its own the EU wants to change the dependency to the US by developing its own military and defence capability (Brzozowski, 2019; Juvan & Prebilic,
12 2012). In July 2016, NATO and EU leaders signed a joint declaration strengthening the EU's military role with the help of NATO, because the EU wants NATO to help them with the Russian threat (Brattberg & Valásek, 2019; Keohane, 2018). Member States can choose how they wish to cooperate in the field of defence. This is possible through the EU and NATO, depending on the capability project or operation at hand (Keohane, 2018).
2.1.4 Franco-German relationship
Since the EU is dealing with a combination of more complex security issues and reduced resources, cooperation is very important. Cooperation takes place when Member States are looking in the same direction and have the same priorities (Gilli & Gilli, 2017). The EU is involved in various CSDP missions and operations, making it an important security actor in North Africa and the Middle East where it conducts various training and police missions to combat piracy and terrorism (Daehnhardt, 2018). However, when it comes to relations between the Member States, the process of cooperation is slow. The various security issues and crises such as the Euro crisis, the migration crisis, but also the civil war in Syria cause fragmentation in Europe. In terms of spending on defence and capabilities and in terms of the will to invest in and deploy capabilities, there are huge disparities between Member States (Daehnhardt, 2018; Menon, 2015).
Common understanding, however, is needed for effective cooperation. In the Franco-German defence cooperation this common understanding is limited due to major differences between the two countries. Research has shown that the strategic culture of both countries is the heart of the problem (Daehnhardt, 2018; Dickow, De France, Linnenkamp & Maulny, 2015; Kempin & Kunz, 2017; Kunz, 2018, 2019; Mölling & Maulny, 2020; Pannier & Schmitt, 2014). It is not only a Franco-German problem, other countries in the EU also have different strategic cultures that causes the national level to stay dominant. When no compromise can be made, it endangers the entire European defence as the EU will struggle to improve its military capacity to act (Major & Mölling, 2018). This thesis only focuses on the defence cooperation between France and Germany, which makes it limited.
The Franco-German relationship in defence cooperation has been institutionalized since the Elysée treaty in 1963 (Mawdsley, 2000; Pannier & Schmitt, 2014). Pannier & Schmitt (2014) examined the relation between policy convergence and institutionalized cooperation in order to find out what the prospects for trilateral concord among the UK, France and Germany in
13 defence policies are and to find out if more institutionalized links among them lead to more convergence in the defence policies. British, French and German defence policy doctrines, instruments and practices since the end of the Cold War based on several indicators and the bilateral relations between the states are qualitatively analysed. It becomes clear that France’s policy doctrine has been converging towards British tendencies, rather than German ones. Even though the relationship between France and Germany has been institutionalized, there is no convergence of the defence policies. It turned out that France and Germany had different strategic cultures during the Cold War, so the cooperation only remained a political symbol.
The strategic culture of France is aimed at becoming a world power, also in the military field, as a result of which they strive for the uphold of a national strategic autonomy (Dickow, De France, Linnenkamp & Maulny, 2015; Kunz, 2019). Both Dickow et al. (2015) and Kunz (2019) have identified the factors that enable, but also limit defence cooperation between France and Germany. Dickow et al. (2015) give recommendations about defence cooperation based on these factors on semi-official conversations with actors in the French and German Ministries of Defence and Foreign Affairs. Kunz (2019) describes the relationship between France and Germany in the field of EU defence in her policy brief, focusing on the countries' potential leadership in EU defence. Her conclusion is that leadership is not in sight, which became especially clear from the discussion about European strategic autonomy about which a large number of European countries, including Germany, are sceptical. France and Germany are dealing with deep structural differences, which makes complementarity the best option to ensure cooperation. Strategic autonomy is about the capacity to independently take security policy decisions, plan for and conduct civilian and military operations, and develop and build capabilities that are required for those operations (Kempin & Kunz, 2017).
France already has a prominent position in the field of defence and security. It has a permanent seat at the United Nations Security Council (UNSC) and has a nuclear deterrent (Mawdsley, 2000). Germany's strategic culture has been influenced by the Nazi period (Mawdsley, 2000). Therefore, Germany has a strategic culture of political and militaryrestraint, only engaging in military interventions when it comes to humanitarian, crisis management or stabilization purposes (Daehnhardt, 2018).
In addition to national strategic autonomy, France also strives for EU autonomy. The EU should become less dependent on the US. Cooperation with Germany and other Member States is
14 therefore a priority for France, as they want to make more efforts in the development of defence capabilities. Even though Germany is not opposed to a European defence, it does not necessarily want to operate independently of the US because that relationship was crucial to their own security during the Cold War because of the constant Russian threat (Mawdsley, 2000; Mölling & Maunly, 2020; Kempin & Kunz, 2017). Furthermore, Germany is not a permanent member of the United Nations Security Council (UNSC). For this reason, one of the main aims of Germany was to achieve integration of Germany into NATO (Mawdsley, 2000). Therefore, Germany wants to fulfil NATO's requirements, because NATO is the foundation for European security (Dickow, De France, Linnenkamp & Maulny, 2015; Pannier & Schmitt, 2014).
The aspects that provide a different view of the notion of strategic autonomy are the defence budgets, the role of the army and the use of force, the threat perception, regional priorities and readiness to resort to military action (Kempin & Kunz, 2017; Pannier & Schmitt, 2014). The defence budgets of France and Germany differ mainly because Germany does not possess a nuclear deterrent and is restrained in funding military activities because it does not participate in many military interventions (Pannier & Schmitt, 2014). France, on the other hand, has been engaged in various conflicts with military activities ranging from the training of local forces to high-intensity manoeuvre warfare to peacekeeping operations. Germany only wants to use force for defence purposes or in the context of multilateral operations (Pannier & Schmitt, 2014; Daehnhardt, 2018). By analysing documents, including speeches and initiatives of relevant actors such as Emmanuel Macron and Angela Merkel, and looking into previous studies on the strategic cultures of both France and Germany in particular, regarding the recent further deepening of defence integration; Daehnhardt (2018) explains that European defence cooperation only works if Germany works closely with France and strives for a common strategic culture. It all depends on the willingness of Member States to subordinate national interests to greater defence integration. A smaller defence budget and reluctance in the use of force lead to Germany having armed forces that are less capable than French armed forces because the French have developed a full-spectrum of military capabilities (Pannier & Schmitt, 2014). In addition, it is easier for France to deploy armed forces because the president decides, while in Germany, the Bundestag has to approve any armed intervention of the Bundeswehr (Daenhardt, 2018).
The priority areas to the defence and security of both countries are far apart when it comes to different threat perceptions and geographical location. France is more involved in tackling
15 threats in the South such as in the Mediterranean basin, part of Africa, the Arabo-Persian Gulf and the Indian Ocean. Germany is focusing on the ongoing Russian threat through the threat perceptions of neighbour countries such as Poland and the Baltic countries. An example of a divergence in the priorities is the development of military equipment. France prefers military equipment suitable for desert conditions, whilst Germany wants to produce equipment that is more fitting to use in the European climate (Kunz, 2019; Mawdsley, 2000). Mawdsley (2000) has examined the changing nature of European armaments collaboration in order to assess the level of armaments policy convergence among the three main arms-producing Western European countries: the UK, France and Germany. This is done on the basis of document analysis of official publications, magazine articles, newspapers and speeches from both French and German literature and interviews with representatives from the general defence manufacturers and aerospace industrial associations that enhance the documentary research. Mawdsley (2000) concluded that different national armaments policy cultures, such as strategic cultures and state-defence industry relations affect collaborative projects.
In 1958, France, Germany and Italy decided to work together to develop weapons, mainly focused on the development of a battle tank and a fighter aircraft (Mawdsley, 2000). Germany decided to buy an American Starfighter fighter aircraft in 1958 and develop its own tank in 1963, which made the projects collapse (Mawdsley, 2000). The positions of France and Germany were different, because Germany just wanted to improve their own tank Leopard 2, while France wanted to develop a whole new tank that was less armed (Mawdsley, 2000). A later attempt to jointly develop a tank did not succeed as well because Germany did not want to invest. The production of the Transall military transport aircraft was a partial success since it was actually produced only more towards the French ideal. France needed an aircraft for their commitments in North Africa, while Germany wanted to develop an aircraft that could be used in Central European weather conditions (Mawdsley, 2000). It has improved the situation for aerospace industries in both countries. There were other successful projects, such as the joint development of the Tiger helicopter, anti-tank and surface-to-air missiles (Kocs, 1995). There was also more joint military research, which provided more technical data for future projects.
In addition to the fact that France and Germany have different strategic cultures, both countries also differ in their position towards the defence industry. Germany and France are the largest arms exporters in the world and are good for about 40 percent of the defence industry in Western and Central Europe (Major & Mölling, 2018; Maulny & Mölling, 2020). According to Major
16 and Mölling (2018), there are risks to European security if France and Germany do not compromise in the defence and security industry. France and Germany differ enormously in their approaches to defence in many areas, which can hinder European security. The study by Maulny and Mölling (2020) identified the problems that the potential has to hamper or even prevent a new defence industrial cooperation among France and Germany. The potential problems have been described by means of a document analysis. On the basis of this, the researchers entered into a discussion with representatives from industry and with government officials from the relevant ministries. Ultimately, three major problem areas were identified, namely the political / strategic framework, the defence export regulations and the defence industrial policies. France and Germany agree that things need to be done to make EU defence cooperation work (Major & Mölling, 2018; Maulny & Mölling, 2020).
The defence industry is an important employer for regional economies in both countries which can at times influence security and defence considerations (Kempin & Kunz, 2017). Kempin and Kunz (2017) have examined how France and Germany can contribute to achieving the goal of European strategic autonomy. In their report they describe the differences between the two countries on three different dimensions of strategic autonomy, namely political, operational and industrial autonomy. Based on the description of the differences, they provide recommendations that are suitable to move beyond the present state of affairs in order to reach a new level of cooperation between both countries. They conclude that France and Germany have different strategic cultures and complementarity is the only thing that can ensure better and more effective cooperation. It is therefore important to focus on the strengths of both countries and not on the differences. In addition, it is important that France and Germany also get the other European partners on board as they show the way for cooperation at the European level. The defence industries in both countries are however, structured differently. France's defence industrial policy focuses on the development of strategic autonomy. Companies that produce military capabilities are state owned, eensuring the state’s influence on the defence market. In Germany the defence industry is a key aspect of the economy. Its defence industry mostly consists of small and medium free market enterprises, more bound to rules and regulation than the French production companies (Maulny & Mölling, 2020). A major stumbling block is the discussion about joint exports of military products. The EU depends on exports outside Europe, as the European market is too small. European defence industries consequently compete with each other within the EU and on global markets (Kempin & Kunz, 2017). Germany is very reluctant in exporting military products, resulting in the French industry
17 and politics to be unwilling to invest in joint projects. Why develop something if you are not going to export it? (Major & Mölling, 2018).
Since 2014, several cooperation projects have been carried out between the two countries, including the development of a joint indirect fire artillery system, a new major ground combat system and further development of the Eurodrone program. However, PESCO has had a bad start as the German and French have a different perception of the initiative. Germany wanted to include as many member states as possible in the framework, while France only wanted the most capable member states (Daehnhardt, 2018; Marrone, 2017). Marrone (2017) describes the development of a permanent structured cooperation in Europe. The different positions are mainly due to the above mentioned different strategic cultures. Germany takes a more multilateralist approach, while France thinks that strategic autonomy can only be accomplished by a small group of states that have more capabilities to steer military operations in the right direction (Daehnhardt, 2018; Major & Mölling, 2018).
In January 2019, on the 56th anniversary of the Elysée Treaty, France and Germany signed the new Treaty of Aachen which pledges enhanced cooperation on European policies and reaffirms their foreign policy, security and military cooperation (Treaty of Aachen, 2019). In October 2019, both France and Germany again confirmed investing more in defence cooperation in order to make it more effective, to reinforce the European in NATO and to create a common strategic culture. By doing so, more attention was given to the PESCO projects. In October 2019, France and Germany also agreed on the development and export of German weapons systems.Whilst France was still selling weapons to Saudi Arabia – weapons which Germany believes have contributed to the war in Yemen – Germany effectively vetoed the export of France-German defence systems. The weapons are no longer exported to countries at war. However, this does limit the development of new tanks, combat jets and drones while exports outside Europe are important as the European market is too small (Major & Mölling, 2018).
This section shows that defence cooperation between France and Germany is difficult due to the difference in strategic cultures. Even though these differences are still apparent, more projects seem to be successfully completed in recent time. Following this recent trend, projects about the development and deployment of AI have a chance in succeeding.
18 2.2 Artificial Intelligence
2.2.1 General AI
Artificial Intelligence (AI) already exists for more than five decades and has been widely applied to areas such as medic, finance, education and transport (Kavanagh, 2019; Wang & Siau, 2018). However, AI is an ill-defined and complex term which results in the absence of a globally accepted single definition (Kavanagh, 2019; Sökmen Alaca, 2019). In 1956, John McCarthy and his colleagues defined AI as “making a machine behave in ways that would be called intelligent if a human were so behaving.” More recently, AI has been described as a form of technology which builds computers and machines that can perform actions that requires human characteristics, most notably that of intelligence (Burton & Soare, 2019; Franke, 2019). Performing complex tasks, being able to learn and improve operationally over time, and doing this without human input, refer to the intelligent aspect of machines (Burton & Soare, 2019). However, intelligent behaviour is a broad term and open to discussion. According to Cummings et al. (2018), a house thermostat is intelligent according to these definitions, as it senses and adjusts the temperature. This is certainly different from self-driving cars without any human control, or a drone that selects its targets without human control.
AI is a General-Purpose Technology (GPT), which are known to create new markets, industrial changes and economic and social environments (Brynjolfsson & McAfee, 2017; Dafoe, 2018; Foray, David & Hall, 2019). The reason AI is a GPT is because AI automates capabilities that humans cannot automate and devices with machine learning abilities learn faster and develop special skills more efficiently than humans are able to do (Brynjolfsson & McAfee, 2017). AI has the aim to support, substitute, and improve human performance in tasks such as decision-making and prediction (Maas, 2019).
In order to participate in the current information technology revolution centred on AI technology, societies will have to adapt to AI (Eetgerink, 2019). To better understand the impact of AI on society, it is important to outline the definition of AI. AI is an umbrella term that refers to various disciplines and techniques such as machine learning and deep learning (Wang & Siau, 2018). Although the concepts in the media are used interchangeably, there is a certain hierarchy where AI is the overarching technology.
Moving beyond the basic definition that is mentioned above, AI can be divided into three categories; narrow AI, general AI and super intelligence. Narrow AI refers to the specialization
19 of AI machines, which means that the machines can perform a specific task by using machine learning and deep-learning tools with humans providing the input (Burton & Soare, 2019; Carriço, 2018; Eetgerink, 2019). In this context, the technology cannot use or migrate the knowledge or behaviours it taught or has learned in one context to another (Burton & Soare, 2019). The technology can only perform a single task at a time. For example, Google’s search engine is used to quickly answer factual questions which would be difficult for humans to answer, however the technology is not capable of using that information systemically and find the connection with other things (Carriço, 2018). However, general AI can perform tasks that are similar to the cognitive intelligence of human beings (Burton & Soare, 2019; Carriço, 2018). In addition to performing the task it was designed to perform, general AI can understand context and intentions and assign information and behaviour to a specific context. In this way the machine can adapt to a certain situation (Burton & Soare, 2019). Examples of this type of AI are robots that are shown in science-fiction films, such as Star Wars, since this form does not yet exist in reality. The last category is artificial superintelligence, where the machine is smarter than the smartest in practically every field, including social skills, general wisdom and scientific creativity (Carriço, 2018; Horowitz, 2018).
The most important advances in AI are currently being made through machine learning techniques, particularly deep learning (Franke, 2019). Machine Learning is a subfield of AI, which gets a prominent position in the general debate on AI (Eetgerink, 2019). In order for the machine to obtain logical and rational results with the inputs entered as data, an algorithm is needed which is a set of unambiguous instructions that a mechanical computer can manage (Sökmen Alaca, 2019). In this way the machine can improve its performance without the explanation of humans on how to accomplice all the tasks (Brynjolfsson & McAfee, 2017). Machine learning also has a subfield, called deep learning. Deep learning refers to the self-learning of the algorithm and enables AI and machine self-learning technology to learn and improve its performance (Eetgerink, 2019; Sökmen Alaca, 2019). So, AI is the umbrella term where machine learning and deep learning both are part of.
20
Figure 1. Concept of AI
2.2.2 Military AI
Much AI technology is also known as a dual-use technology, which means that the specifications ensure that the technology can be used in a military or civil setting (Burton & Soare, 2019). Where it brings many benefits in civil fields, it provides fears and worries in the military field (Sökmen Alaca, 2019). According to military professionals, experts and strategists, AI is increasingly being used in and developed for the military realm (Allen & Chan, 2017; Scharre, 2018). Indeed, there are already weapons-guidance systems that make decisions independently of human input and intelligence agencies that use algorithms to identify patterns in large data sets (Garcia, 2019). Garcia (2019) describes the challenges that the militarization of AI entails beyond the realm of security itself, such as geographical and gender balance. Above all, she makes clear that little has been written on the subject and that much more research needs to be done.
The role of military AI will dominantly impact the tactical domain, but will have important strategic implications. Military AI ensures that processes can be executed faster, more efficiently and cheaper because machines are faster than humans in analysing and making decision based on big data (Payne, 2018). Military AI capabilities are used to localize, track and target various weapon-systems of the enemy through which an attack can be executed. Having a more advanced process than your adversary improves offensive mobility and makes you harder to hit (Payne, 2018). Because the attacks occur from a distance, the attack often
21 comes as a surprise, so the opponents cannot react quickly. AI makes it possible to analyse dynamic battlefield conditions in real time and strike quickly and optimally while minimizing risks to one’s own forces (Davis, 2019). The use of military AI is therefore attractive to militaries for both defensive and offensive purposes.
The public debate about military AI is mainly focused on lethal autonomous weapons (LAWS), the so-called ‘killer-robots’, which performs critical functions in the targeting cycle of a military operation without human intervention (Bode & Huelss, 2019). Examples of lethal autonomous weapons are swarms, autonomous drones or underwater vehicles (Burton & Soare, 2019). These systems are only used for specific missions, but activists are currently lobbying for a pre-emptive ban on the development of those systems (Franke, 2019). On the one hand, autonomous weapons provide benefits such as reducing military expenditure. On the other hand, critics believe that the systems are incapable of distinguishing between combatants and civilians, making the use of those systems in the battlefield illegal. Next to that, it is not clear who is responsible for the actions of the AWS (Altmann & Sauer, 2017). Fully autonomous weapons must of course be distinguished from other weapon systems that are not prohibited by Human Rights Watch. These are weapon systems that only engage until a human confirms the selected target and weapon systems in which a person does have influence, but when the person does not respond, the weapon system carries out the attack by itself (Schmitt & Thurnher, 2013). AWS are, however, not the only way AI is used in the military realm. AI military applications have a wide-ranging scope that affects all five domains of warfare (land, sea, air, outer space, and cyberspace). In addition, AI also affects intelligence, surveillance, command and control, communications, and reconnaissance especially through machine learning (Allen & Chan, 2018; Garcia, 2019).
At the operational level, military AI can be used in various ways. Nowadays a huge influx of intelligence from many different sources makes it important to effectively categorize and analyse this information in order to improve decision-making. This can be made possible by using machine learning, which ensures better situational awareness (Davis,2019). An example is the Project Maven where The Pentagon partnered with big technology firms in order to categorize video footage collected by drones to identify hostile activity in the terrain of Islamic State (Davis, 2019; Franke, 2019). Military AI is also used in the context of situational awareness for image and face recognition, speech recognition and translation, the geolocation of images, and pattern-of-life analysis (Franke, 2019).
22 Autonomous vehicles are an example of a military AI application which is already used in various military operations. Autonomous vehicles provide navigation through a hostile terrain. This can also be used through swarms, a large group of unmanned devices who act independently but coordinated in order to discover an environment (Scharre, 2018). Eventually, the systems can report certain changes that can bring benefits in the battlefield (Davis, 2019). Military AI applications are also being used to support the logistic processes in the military, such as the maintenance of aircrafts. By predicting the maintenance of an aircraft, the process becomes more efficient. Next to that, AI is also used in in wargames and battlefield simulations in order to learn ‘wargamers’ what the effect of dynamic conditions, such as weapons, allies and interventions, have on the outcomes and decision-making process (Davis, 2019). AI capabilities also have an important role in cyber operations, since the cyber realm is a digital domain. By using AI, computer networks can be mapped and hacked.
The potential of weaponization is probably the biggest threat from AI. Although it is not yet clear how AI has an effect on the military, the increasing development of military AI applications and semi-autonomous drones that can operate without human intervention lead to concerns among scholars and experts (Dahab, 2018; Franke, 2019). Just like other technological innovations, whether the machine gun, the railroad or nuclear weapons, AI could shake up the balance of power and international conflict. One of the most cited statements in the work about the development and deployment of AI is the statement of President Vladimir Putin. According to him, ‘[w]hoever becomes the leader in this sphere will become the ruler of the world’ (Allen & Husain 2017, Carriço, 2018). Russia, China, the US, Israel, France and the United Kingdom are developing or using military AI and plan to invest even more (Carriço, 2018).
2.2.3 AI in Europe
In 2016, Europe raised awareness to the need for global rules on AI in the global strategy for foreign and security policy (European External Action, 2016 in Carriço, 2018). The European Council, The European Commission and the European Parliament named the importance of AI in various meetings and reports in 2017 (Council of the EU 2017a; Council of the EU 2017b; Council of the EU, 2017c; European Commission, 2017b; European Parliament, 2017 in Carriço). The Council of Europe has urged the Commission to create a European approach to AI by early 2018 (Council of the EU, 2017). This approach makes clear that the EU is following a human-centric approach to all domains of AI, including military AI, since the EU wants to
23 leave people in control of the machines (Carriço, 2018). Carriço (2018) analysed the potential benefits and drawbacks of AI in all domains for the EU and provides a human-centred perspective on the topic. In his article, Carriço (2018) describes the opportunities and threats by non-military and military AI. He gives policy recommendations that describe how the EU can participate in the AI race. If the EU wants to become bigger in non-military and military AI development, according to Carriço (2018), it must have a human-centred perspective, since humans should not be replaced by AI robots, since they cannot sense things like emotions and intentions.
In April 2018, a declaration was signed on overall AI collaboration by 25 EU countries including the UK (Wu, 2020). Throughout Europe, member states are developing a national AI strategy or have already published an AI strategy (Franke, 2019). The European Commission asked EU member states to publish the strategies or programs by mid-2019 (Coordinated Plan on Artificial Intelligence). Countries such as Belgium, Denmark, France, Finland, Italy, Sweden, and the UK have already published a strategy. Member States would like to compete in the field of AI, but to compete against superpowers such as China and the US, EU-cooperation is important (Wu, 2020). Wu (2020) has studied how to make use of the characteristics of AI to better protect and enhance information security by describing the legislation and planning on this topic in Europe and the US. AI solves many information security options by improving the efficiency of certain processes such as face recognition and real-time viable intelligence. However, it can also pose a threat such as disinformation.
In December 2018, the European Commission released a Coordinated Plan for Artificial Intelligence, which is the coordinated AI strategy for the EU member states in which actions about the development and deployment of AI are proposed for the coming years. Shortly after, a draft on ethical guidelines for trustworthy AI was published by the European Commission's High-Level Expert Group, which provides policy and investment recommendations. During the rise of AI in Europe, the EU started looking at ways to regulate AI, as the EU would lose the oversight of large technology firms. In 2018, the EU announced its data privacy rules, the so-called General Data Protection Regulation (GDPR). These regulations serve as a benchmark for tougher regulation in other parts of the world. In the development of AI, the EU values substantiated rules that ensure the protection of fundamental human rights.
24 In February 2020, the EU released a white paper on some non-military areas of AI. The focus is mainly on improving and promoting EU's skills and industrial capabilities in various economic categories. In addition, as has already been shown, there is also a focus on the regulation of data. The EU has not included military purposes or AI in the strategy. This absence however, does not mean the subject is not a topic of discussion in the EU
On 12 June 2019, a conference on AI and EU security and defence was organized by the EU Institute for Security Studies (EUISS), The Estonian Ministry of Defence and the Permanent Representation of Estonia to the EU, the Ministry of Defence of Finland and the Permanent Representation of Finland to the EU (European Commission, 2020). Here, various experts from EU institutions, governments, academia and industry discussed AI and the impact on the defence sector. EU-cooperation is also important in the context of AI, because AI military capabilities present both opportunities and challenges (Hill, 2020). Hill (2020) took a closer look at the state of discussion around AI-based military applications within NATO by framing the overall strategic context and by describing the publicly-accessible work that has taken place within NATO on AI issues. Individual NATO allies are developing their own national strategies for military applications or AI, but this creates a lot of difference in those frameworks. A multinational setting would help improve AI-enabled military applications and their implementation, allowing NATO to play a key role in this. Therefore, a Food for Thought paper on military AI has been publishedin which the need for military AI capabilities in Europe is addressed (Food for Thought Paper, 2019).
Next to the Food for Thought paper, the EDA (2020) launched a blueprint about AI innovation. The EDA aims to create a common understanding of AI terminology to enable cross-border collaboration, as discussions with Member State experts show that there are divergent interpretations about what AI actually means (EDA, 2020). A common understanding is important for Europe's strategic autonomy. In addition, the EDA will invest in the identification and analysis of military AI applications. Finally, the EDA of AI action plan can be drawn up, which clarifies how Member States can work together concerning the development of military AI, which will be formally validated at the end of 2020 (EDA, 2020).
Although there have been some developments of AI in Europe, as mentioned in the paragraphs above, little research has actually been done on the military aspects of AI in European context. In the EU Member States, the development of AI and robotics is mainly focused on the civilian
25 dimension of AI, such as the industrial sector, not the military dimension (Franke, 2019; Haner & Garcia, 2019). The article of Haner and Garcia (2019) describes the future impacts that increasingly autonomous weapons could have on international security. It also ranks the top five world leaders according to their intent to develop AWS hardware and their level of AI expertise and highlights the importance of ongoing efforts to restrict or ban the use of AWS. This study points out that oversight of increased autonomy in warfare is important since deadly technology can also be used by terrorists and other people who want to take advantage of it. The top competitors in the AI arms race are the US, China, Russia, South Korea and the EU.
Several Member States such as France, Germany, the UK, Sweden and Italy are already developing autonomous military robotic systems, while other Member States are still lagging behind or even want to prohibit the use of AWS (Boulanin & Verbruggen, 2017; Franke, 2019; Haner & Garcia, 2019). Therefore, military AI cooperation is still limited. Franke (2019) conducted a comparative case study on the European thinking on AI in the military and how European armed forces plan to use AI by mapping and assessing the approaches that Germany, France and the UK are taking to using AI in the military. She analysed official documents, information gathered through personal conversations and interviews and reports on industry projects. The comparative study showed that the defence industries are developing AI capabilities in all three countries. France and Germany are, however, at the opposite ends of the AI spectrum in the EU. The UK is in between both countries. France sees a lot of potential in military AI, while Germany has little regard for the military aspect of AI. In Germany, officials find it difficult to raise this issue and the focus is only on whether and how to ban "killer robots" or lethal autonomous weapons systems (Franke, 2019). When European countries do not change their position in relation to military AI, defence cooperation becomes more difficult. The study has already provided more insight into the different approaches of countries in the field of military AI particular, but this thesis examines whether the different approaches Are equal to other approaches towards military cooperation in the context of the EU.
Recently, the EU has shown progress in the development and deployment of AI. There are a number of projects developing AI-enabled European military systems, such as BAE Taranis, Dassault nEUROn, Airbus and Dassault Future Combat Air System, BAE Tempest and Artemis. In addition, the EU has also taken the first steps in the development of military AI in collaboration with NATO (Haner & Garcia, 2019).
26 The aim of this thesis is to investigate whether the different positions of France and Germany in military AI have been formed in the same way as in the development and deployment of other military capabilities.
27 3. Methodology
In order to answer the question what the hurdles for the EU-cooperation of the development and deployment of military AI capabilities are, a comparative research is carried out of the countries that are the motor of European integration. A qualitative approach is used, analysing data derived from domestic policies from France and Germany, such as strategies and policy papers, data from journal articles, newspapers, and interviews with relevant actors. Besides the two Food for Thought paper (2019) and the blueprint of the EDA (2020), there are no EU documents to analyse for this thesis. The two EU documents do not provide insight into the hurdles for EU defence cooperation in the development and deployment of military AI between France and Germany.
France and Germany are key European states, in terms of both economic strength and political and military power (Pannier & Schmitt, 2014). Both are members of the same international institutions such as the EU, NATO and the UN, but have different domestic political institutions and historical experiences. As discussed in the previous chapter, post-Brexit European defence depends on cooperation between France and Germany. The countries are responsible for almost 50 percent of the defence and industrial capabilities. it is examined whether the French and German positions on military AI differ the same way as previous military cooperation issues, and by doing so limiting military AI cooperation in the EU. It is expected that deep-rooted differences between France and Germany, such as their strategic cultures, in the development and deployment of military capabilities are still present, also in the field of military AI. In order to examine this, a qualitative research is being conducted. By analysing the motives and strategies of France and Germany as outlined in their military AI strategies or positions, the differences that can hinder the defence cooperation in the development and deployment of military AI become clear and linking them to factors that previously resulted in unsuccessful cooperation the documents are analysed.
3.1 Operationalization
The documents are analysed by using predetermined categories. The categories are based on the factors that impacted previous defence cooperation between France and Germany in the field of military capabilities which either enabled or prohibited cooperation. On the basis of the literature on Franco-German defence cooperation in the development and deployment of military capabilities, the countries appeared to differ from each other on several factors. Concerning EU cooperation, this thesis focuses on the cooperation between states in order to
28 achieve a common ground, specifically in the field of defence. Indicators for this factor are cooperation in projects or interventions of organizations like PESCO. Concerning attitudes towards the international security field, this research examines whether the country is focused on multilateralism or strategic autonomy. In order to analyse if a country is more focused on multilateralism there has to be a diplomatic interaction that requires more than three states to follow international norms and pay more respect to international institutions. An indicator of this would be the relationship with NATO. In order to mark a country as focusing on strategic autonomy, the state has the ambition to independently take its own decisions in foreign policy and security and not only obeying to rules set out by others, such as the United States, Russia or China. This can be reflected in the role of the armed forces, the threat perceptions, defence budgets and regional priorities, both geographical as well as economic priorities. Finally, for the defence industry, I focus on the way the countries implement the industry policies.
3.2 Data collection
In order to examine the position from both France and Germany when it comes to military AI this thesis will use data that is derived from domestic policies from France and Germany, such as strategies and policy papers, and data from journal articles, newspapers, and interviews with relevant actors The data is collected from official government publication websites1, large newspapers, and online defence and technology news forums. Relevant actors in this thesis are people in political positions who may influence the development and deployment of military AI, such as the country's leaders and defence ministers. In addition, actors in important positions in the military and experts on AI, specifically military AI, are relevant actors for this research. They provide insight into the working method of the military and the opportunities of military AI for the countries. Key words that were used to search for these documents are military artificial intelligence, development, deployment, defence industry, national AI strategy, military AI strategy, (L)AWS, weaponization of artificial intelligence. Since this thesis focuses on the differences between France and Germany, it is a requirement that the documents entail information on one or both countries. The included documents must contain information about the development and deployment of military AI or the possibilities for the countries. Documents that contain the positions of relevant actors in the field of the development and deployment of military AI are also included.
1 Military AI. France: https://www.gouvernement.fr/en/defence-and-security Germany: https://www.bmvg.de/en
Non-Military AI. France: https://www.gouvernement.fr/en/france-top-documents Germany: https://www.bundestag.de/en/documents
29 The official documents examined in this thesis are the national AI strategies of France and Germany, which set out the country's plans for AI development and deployment (The Federal Government, 2018; Villani, 2018). The documents cover not only military AI, but also AI in other areas. In addition, both countries have issued a policy paper that highlights the (possible) strategy in the field of military AI (Ministère des Armées, 2019; Position Paper of the German Army, 2019). In addition to the official documentation, several statements of and interviews with relevant actors are examined. The relevant actors in this case are German Chancellor Angela Merkel, French President Emmanuel Macron, a German lieutenant and general, the French defence minister and the head of the intelligence analysis department of the Bundeswehr (Buck, 2019; Bundesministerium der Verteidigung, 2018; Dahlmann & Dickow, 2019; DW Made for Minds, 31 January 2020; Federal Academy for Security Policy, 2018; Freist, 2018; Gouvernement, 2019; Leinhos, 2020; Köver, 2019; Parly, 2019 April 5, December 12; Reuters, 2018; Statement of France and Germany, 2018; Thompson, 2018). The documents discuss the motives and reasons for developing or not developing military AI. In addition, the documents also clarify the positions on lethal autonomous weapons of France and Germany. By not only examining the official documents, but also including statements and interviews from relevant actors, a broader view is taken of the different positions of France and Germany. For the analysis, the documents will fully be examined to discover the rationale behind certain statements and strategies.
3.3 Limitations
This thesis has several limitations. First of all, the number of official documents examined in this study is limited, as there is still not much written about military AI and publications about military policies and strategies at national and European level are often not accessible. Germany is probably involved in military AI, but this might not have been publicly published, which means that this thesis can give a distorted picture. Insight into the possible strategy of the Ministry of Defence in Germany could provide more insight. In addition, there are no documents available at European level that provide insight into the defence cooperation on military AI and the ongoing PESCO projects, so this could not be analysed. The documents that were analysed differed in terms of the format, audience and language because some of the documents were official policy documents from both countries and the other documents were retrieved from newspapers and online news forums. Those articles are often intended to make a good story, and may have a bias due to a particular background a newspaper or online forum
30 has. Some of the unofficial documents were only available in German which can influence the interpretation of the text. In this thesis the documents were analysed by one researcher, which reduces reliability. In addition, the analyse focuses specifically on the development and deployment military AI in France and Germany, which lead to problems with generalization because it is not applicable to other member states or contexts. This reduces the external validity. There may also be other aspects that create a difference between countries and make EU-cooperation more difficult, but I will focus on the factors emerging from previous research on the Franco-German relationship. Since not the entire concept is measured, the validity decreases. Other research methods such as interviews and more extensive use of primary data would provide a better explanation of the findings.
