The Estonian defence leadership recognises Artificial Intelligence (AI) as an increasingly important suite of technologies that will transform society, the economy, and the defence sector. There is a clear understanding that Estonia must remain part of this technology’s wave, not least because its defence forces will have to interact and remain interoperable with the allies racing to embrace AI-enabled capabilities.

Concurrently, Estonia’s national defence development faces a major challenge of addressing significant capability gaps within the compressed timeframes, given the assessments that the reconstituted Russia’s capabilities will pose a direct and existential threat to Estonia by the end of the decade, if not earlier. Combined with resource constraints and inherent scepticism within the military about emerging technologies that so far hold more promise than they deliver, this rapid capability build-up is pushing investments into defence AI development down the list of priorities.

On the other hand, war in Ukraine is increasingly supplying insights into what effects the use of AI can have in a battlespace when combined with old technologies and pique interest in how small states such as Estonia could employ this technology to offset imbalances vis-à-vis a numerically superior enemy force. Some in-house development projects aimed at digitalising “kill webs;” enabling better information sharing and enhancing common battlespace awareness are also emerging as important vehicles for introducing AI.

At the same time, the Estonian security, defence, and space industry, dominated by agile start-ups and small or medium enterprises, is emerging as a significant driver of defence AI development. Enterprises in Estonia are quickly becoming important sources of concepts of AI applications in defence and security, even though their products and services based on those concepts are often likely to reach foreign customers sooner than the Estonian ones.

Most importantly, a wave of procurement of state-of-the-art weapon systems and equipment is bringing the Estonian military into close contact with advanced technologies that will include, as part of the package, elements of AI, further necessitating increasing knowledge and competence in AI technology. In this regard, Estonia will have to strengthen its military training and education system and find ways to leverage conscription and reserve training more effectively to enhance AI competence.

1 Thinking About Defence AI

As for many nations, Estonia’s journey into defence applications of AI started in the civilian sector, where its main strengths lie in information technology, digitalisation of public services, cybersecurity and, increasingly, semi-autonomous robotics. AI, defined as “a system based on an autonomous software algorithm capable of learning, allowing it to perform tasks that typically require human intelligence” (Government of Estonia 2019), is making its way into civilian applications through these key areas, with the national strategy providing the overall direction for public and private stakeholders. The second iteration of this strategy is being drafted and will give a better conceptual basis, context, and coherence to the overall effort (ERR 2023b). It is important to note, however, that the level of digitalisation—fairly high in various international indices (e.g. EU’s Digital Economy and Society Index or UN’s E-Government Development Index)—may not necessarily be a reliable indicator of the level of digitalisation and AI adoption across the defence sector.

1.1 Estonia’s Strategic Culture of Pragmatism

Estonia’s strategic culture shapes its defence AI approach in important ways (Salu and Männik 2013; Atmante et al. 2019). First and foremost, Estonia relies on NATO and the EU to ensure national security, particularly in managing a persistent, multifaceted, and overwhelming threat Russia poses. This key tenet rests on clear-eyed understanding how large the power asymmetry with Russia is that Estonia needs to offset to survive as a sovereign nation. Reliance, however, is not seen as a one-way street. Rather, it requires both maximising Estonia’s contribution to the collective enterprise wherever possible and being an active and involved member of these organisations. This also entails continuously investing into credible national defence capabilities and allocating the necessary financial resources. Since re-establishing its independence after 1990, Estonia had to rebuild its defence forces from scratch.

Second, Estonia’s security culture reflects the understanding that developing national military defence and further focusing it on homeland (territorial) defence is confined neither to national boundaries nor solely to the defence sector. It requires close interaction (and thus interoperability) with allies and the ability to draw upon inputs from broader society, including the corporate sector and science. Third, its strategic culture is focused on practical solutions that deliver concrete results, preferably within a relatively short time. In general, Estonia shies away from slow-motion conceptualisations and over-strategizing, even though in such sectors as cybersecurity the value of multi-stakeholder strategy-making to bring together all relevant experts and sustain understanding is well appreciated (Kaska et al. 2021).

This pragmatic and practical mindset is important as it shapes how defence AI is approached—with a high degree of uncertainty-induced caution in the military mixed with sufficient space for entrepreneurial attitude on the industry and policy side “to get things done.” It also reflects the military’s general approach to technology adoption, underpinned by a clear understanding of Estonia’s narrow margin of error in initiating (or not initiating) major change and limited resources to pursue any new high-flying ambitions (see Suurkask 2023). Such pragmatism means it cannot and will not be a trailblazer of defence innovation until and unless the evolving nature of military threat from Russia makes a rapid introduction of some unique solution—that no one else in the Alliance has yet adopted at scale—a clear and urgent imperative.

1.2 The Current Defence AI Strategy Void

Estonia still needs to have a dedicated defence AI policy, but the MoD is gradually rising to the challenge of drafting it. The current approach tantamount to informal policy on AI-related issues is shaped through its existing R&D and defence industrial policies, but also by the developments in the framework of NATO, the EU and minilateral initiatives that the MoD is keen to participate in. As a member of NATO, Estonia subscribes to the Alliance’s AI Strategy of 2021, and as a member of the EU, it participates in defence cooperation frameworks that have elements of AI technology and AI-enabled capabilities (e.g. Permanent Structured Cooperation, European Defence Fund, European Defence Agency, etc.). Furthermore, it joined the US-led AI Partnership for Defence initiative—a “coalition of the willing” that reinforces, amplifies, and benefits from the US leadership in the field (Chief Digital and Artificial Intelligence Office 2022). This is seen as recognition of Estonia’s strengths in cybersecurity—a field where synergies with AI are significant.

The MoD’s key strategic message is that AI is pivotal to future military capabilities and that Estonia must keep abreast of the developments in this field and contribute to them in the areas of its technological excellence and expertise. Although the understanding of what capabilities will be transformed or created by AI’s use is still lagging in the wider Estonian defence establishment, the basic principles that the MoD follows in relation to AI projects for defence are (Roundtable, 2 September 2022):

  1. 1.

    They must add value.

  2. 2.

    They must simplify rather than complicate employment of military capabilities.

  3. 3.

    They should be of dual-use purpose.

There is also strong determination at the policymaking level to get Estonia involved in a web of multinational collaborative initiatives that will allow tapping into the results, ensuring interoperability and avoiding early (and often costly) mistakes made by “first movers” in the field. Such involvement also compensates for the lack of critical mass that hampers small nations such as Estonia in their technological and capability development ambitions.

In this regard, two key role models emerge as pivotal for Estonia. First, the United States is a global leader in AI technology and a strategic partner of immense importance for transatlanticist Estonia. In addition, the UK is as another long-standing and trusted strategic partner in military affairs (Jermalavičius and Billon-Galland 2023). The UK is a framework nation for both the Joint Expeditionary Force (JEF) in which Estonia participates alongside other ten nations and NATO’s enhanced Forward Presence (eFP) battlegroup deployed in Estonia. It has recently committed a high-readiness brigade for Estonia’s defence (Estonian MoD 2023b), thus making interoperability with the British forces a key consideration for Estonia. The UK also is among the leading technological powers in the Alliance, further boosting its relevance to Tallinn’s defence AI thinking.

The pivotal role of these two nations does not imply that other partners do not matter. Estonia participates in the French-led European Intervention Initiative and hosts French troops as part of NATO’s eFP, so the evolution of the French defence AI strategy and technology leadership will certainly be of interest to Tallinn. Collaboration with Germany—a source of military technology and equipment of growing importance to Estonia and a nation that resolved to enact far-reaching transformation of its defence as part of the Zeitenwende—will shape Estonia’s defence AI agenda and thinking. Last but not least, some of the like-minded small allies with similar strategic outlooks—such as Finland, Norway, Latvia and Lithuania—are likely to be among partners of choice in keeping up with the defence AI trends in the Alliance.

1.3 Defence AI Ethics: Yes, but…

In terms of general thinking about defence AI, Estonia seeks to subscribe and adhere to the overall principles of its responsible and ethical use, although remaining sceptical of the normative regulation (e.g. arms control treaties in relation to lethal autonomous weapon systems) and advocating instead norms-building approach—something that also defined its approach to international legal norms in cyber domain (Crandall and Allan 2015; Osula 2021).

The government’s thinking—including how it defines autonomous weapon systems and their governance framework—is best reflected in the joint Estonian-Finnish paper presented to the UN expert group’s working session (GGE CCW 2018). In particular, the paper articulates that “autonomy should be understood as a capability to perform the given task(s) in a self-sufficient and self-governing manner” and argues that “humans must retain ultimate control over decisions of life and death.” Crucially, it sets a baseline for the human operator competence in the armed forces. According to the paper, human operators must possess, “at a minimum, an understanding of the performance characteristics of the system and of the operational environment” and, consequently, “if the operator lacks such an understanding, or based on that understanding has no confidence as to compliance with the law, he/she must not permit the weapon to deliver force.”

At the same time, however, some members of Estonia’s government and the business community are concerned that multinational and the EU efforts (e.g. the EU’s AI Act) will stifle development of innovative AI-based solutions in defence and will stall the emerging vibrant ecosystem of AI industry enterprises that are potential suppliers of such solutions (Riigikogu 2023b). Given the focus of Estonia on promoting these ecosystems, restrictive international and EU regulation seems to pose a risk to its future ability to contribute to collective defence AI endeavours. Policy-wise the MoD and other relevant government organisations are bound to continuously balance between nurturing the Estonian AI start-ups ecosystem and responding to the pressures from various directions to constrain the development of technology seen as posing unacceptable level of risk.

1.4 Industrial Push and Military Caution

Estonia’s defence, security, and space industry has indeed been one of the key drivers of defence applications of AI in Estonia—not only through adoption of AI in cyber defence, but also in developing unmanned systems, counter-UAV capabilities, and digital battlefield management solutions. It has also been, thanks to the industry-funded studies and concept development and experimentation (CCD&E) efforts, at the forefront of thinking about the future of warfare where AI-enabled autonomous military systems are dominant in the battlespace (Interview, 20 December 2022). Some of the enterprises working with AI-enabled solutions are in direct contact with the armed forces and industrial partners in Ukraine, seeking to harness their insights and experience from the ongoing war and leverage them towards new concepts and product development.

The military’s visionary thinking has been limited by the realities of defence development, whereby the Estonian Defence Forces (EDF) has had to address most basic capability gaps and needs first, before turning their eyes to more advanced capabilities and emerging disruptive technologies. This pivot is at its infancy but already underway, with parts of the military establishment already considering future scenarios—building upon NATO’s work such as NATO Warfighting Capstone Concept (NWCC). They are also alarmed by the developments in the technology sector of the main adversary, Russia, and inspired by certain elements of war in Ukraine. For these military thinkers, AI will be a major part of capabilities in most of the military functions—from ISTAR (Intelligence, Surveillance, Target Acquisition, and Reconnaissance) to indirect fire support and armoured manoeuvre to combat service support and military administration—in the coming decades (Interview, 14 July 2023).

In line with most of the thinking among Western militaries, AI is viewed as an enabler of dense “kill webs” characterised by compressed OODA (Observe, Orient, Decide, and Act) loops and associated with greater effects on the battlefield (Interview, 21 December 2022). In some of the most forward-leaning thinking, compartmentalised parts of the future battlespace (e.g., absorbing and delaying first waves/echelons of assault by the enemy along particular axes of advance) would be fully unmanned, even though always with human operators “on the loop” (Allik et al. 2021). Most relevant for small nations like Estonia, AI-aided capabilities such as unmanned systems are regarded as a means to limiting manpower losses that even a reserve-based wartime organisation such as the EDF would struggle with, and, furthermore, as a compensator for the lack of human resources that various military functions (e.g., staff work) require (Interview, 14 July 2023).

The military, however, remains cautious and reserved about AI-enabled capabilities and does not have an overarching vision for it. They readily admit that AI is already creating havoc and opening new opportunities in the information warfare domain. However, this is regarded as affecting the political layer of defence more than the military directly and therefore the understanding and management of the associated risks should be a matter for the political and policy-level attention in the first place. In general, they remark that AI will first transform societies before it transforms defence and that the latter transformation will not be in full swing in the coming decade or so anyway (Interview, 6 July 2023). Such remarks justify the focus of acting on the challenges of the more immediate future such as building greater mass, firepower, and reserves. Some military planners contend that, due to technological developments, some form or degree of AI-enabled man-machine teaming might become possible by the end of this decade (Interview, 14 July 2023). However, the overall sentiment is that the entire field of emerging disruptive technology is too unpredictable beyond a time horizon of 5–10 years.

Still, at the most senior levels of defence command, there is already an understanding that something profound is afoot. In some regards, the Allied presence in Estonia is a catalyst of a gradual change in perceptions of the military. The protracted period of experimenting with various iterations of indigenous Estonian platforms unaccompanied by any serious conceptual work on the military side about its potential seems to have created a blind spot in the perspective of the Estonian military. However, the British forces bringing unmanned ground vehicles (UGV) to the exercises in Estonia served as an “eye-opener” about the pace and potential impact of such systems on the capabilities of Estonia’s key allies (Interview, 9 May 2023). Given the importance of maintaining interoperability with these allies, this is a major consideration for envisioning Estonia’s future capabilities and the role of AI in them.

Likewise, Russia’s war against Ukraine is serving as a source of insights about elements of warfare based on the adoption of various new technologies (including commercial-of-the-shelf) that is prompting some progressive thinking—even though data from primary sources on the use of these technologies by Ukraine remains very limited and, according to one senior Estonian officer, does not even provide enough material for lessons identified, let alone learned (Interview, 6 July 2023). Nonetheless, the war in Ukraine is accelerating ongoing efforts to prioritise defence capabilities like ISTAR, that requires greater reliance on ubiquitous and interconnected sensors, or digitised command and control (C2) processes to accelerate information sharing and decision-making. Long-range stand-off fires integrated with comprehensive sensor webs also receive greater attention in the Estonian military thinking, as does the role of loitering munitions (Jäärats 2022).

At the same time this view will be tempered by the observation made at the highest command level that there is no single “silver bullet” in the war in Ukraine, despite some advanced weapon systems being “elevated to the podium” at various points in time. Rather it is the combined, coordinated and sustained use of different systems on a large scale that is seen as most impactful in delivering the desired effects (Jäärats 2022). As a consequence, it is argued that Estonia should invest in more armour, mass and firepower (including long-range stand-off munitions), while adopting mature technologies (including, presumably, AI-enabled) that make their application more efficient and effective.

Future efforts to develop AI-enabled military capabilities will also have to contend with a high degree of caution in Estonian society. A public opinion survey conducted in late 2021 revealed that 62% of respondents agreed that AI can become more dangerous than nuclear weapons, while 85% did not trust the application of autonomy in military capabilities and 89% were against delegating “life or death” decisions to autonomous military systems (Idarand 2023). Partly reflecting this sentiment, but also to meet high standards sought at the EU level for the EU-funded projects, some companies have been instituting clear and robust ethical AI governance policies (Milrem Robotics Undated). They reflect the current state of thinking in Estonia about corporate responsibility and accountability in developing and fielding defence AI applications.

2 Developing Defence AI

Developing AI-enabled solutions is not consciously prioritised by the Estonian MoD when selecting and supporting R&D projects. For instance, in the framework of the defence industrial policy, the MoD provides annual grants to enterprises for developing solutions that are highly innovative, contribute to enhancing capabilities, and have export potential. Some of those successful projects may indeed contain elements of AI and, with time, are more likely to do so given the overall trajectory of technological development in the commercial sector. However, there is currently no deliberate demand pull from the defence organisation to do so. There is some expectation though that the number of such projects will grow and, at some point, will organically reach critical mass (Roundtable, 2 September 2022).

This “bottom up” approach at the policy level is echoed by military practice. The EDF is quite open to testing new solutions such as robotic platforms with various units in the field and providing feedback to the developers. However, it does not place any special emphasis on those solutions being AI-aided or -enabled. The focus is firmly set on whether they resolve any specific technical, tactical, or operational problem and address an existing capability gap rather than on laying ground for introducing new paradigms and concepts of future warfare. The capacity of the EDF to define generic requirements for future capabilities and develop concepts for their employment remains very limited, further constraining the demand pull for AI-based solutions.

This is not to say that the ground is not being laid for such solutions. As an example, the efforts to stand up the Estonian Division (ESTDIV)—a formation that will serve, among other things, as a “plug and play” environment for NATO Allies deploying to Estonia—require addressing various shortcomings in the C4I (Command, Control, Computers, Communications, and Intelligence) systems. The ongoing integration and further digitisation of command and the resource management processes are a prerequisite for future AI-based applications, including those of the allies whose units will become part of the ESTDIV. This will have to overcome the proclivity of national contingents contributed by various Allies to limit information sharing across national lines—a procedural and policy rather than a technical problem requiring the constant use of staff liaison officers (Briefing, 28 September 2023).

In addition, the EDF has been making steady progress with two in-house development projects that form the basis and core of its digitalised battlespace management:

  • KOLT (Kaitseväe olukorra ja lahinguteadlikkuse süsteem)

KOLT is a Defence Forces Situational and Combat Awareness System that was initiated in 2014–2015 by conscripts and reservists with IT background and has since become an EDF-wide solution used by various units and tested during major exercises such as Kevadtorm (Spring Storm) (EDF 2019). Some small development projects to use AI to facilitate its various functions (e.g. mapping, handwriting recognition, etc.) are being pursued by the EDF, showing how a major technological step forward opens opportunities for more AI-related “bottom up” ideas and add-ons.

  • TOORU

TOORU is a fire support system that seeks to combine fire managers, fire control centres, calculation points, weapon systems, fire support officers, logistical components, and aerial fire control elements into a single digitalised “kill web.” Conceived as one of the modules of KOLT, it already generates datasets from various fire systems and is envisaged as an excellent platform for introducing AI solutions aiming to reduce the workload of human operators in preparing fire missions (e.g. performing calculations) (Dieves 2021). Thus, it has already been designed from the start as a system that will both generate data for training AI algorithms and integrate new capabilities based on AI.

Overall, however, there is a sense that the defence organisation needs take a step back and assess the potential and requirements for AI across a range of capabilities. An MoD-funded study project by the National Defence Academy (NDA), together with Germany’s Fraunhofer Institute for Communication, Information Processing and Ergonomics (FKIE), aims to examine the potential requirements and narrow down the broad spectrum of AI applications to more specific priority areas (Interview, 9 May 2023). This should lay the ground for stronger demand pull and the defence AI strategy with a degree of top-down guidance as well as facilitate greater coherence in selecting, encouraging, and supporting development projects in the future.

Estonian enterprises take a lead not only in thinking about, but also developing AI-enabled solutions, often in close collaboration with the University of Tartu and Tallinn University of Technology (TalTech), the country’s two leading universities, and the NDA. A vibrant innovation ecosystem has helped set up several companies that also develop defence products with AI applications (some of them are R&D-intensive in their business model as well) such as Milrem Robotics (UGV development), Threod Systems (UAVs), DefSecIntel Solutions (sensors and surveillance systems), SensusQ (data sharing and intelligence management platforms), Marduk Technologies (counter-UAV solutions), Rantelon (electronic warfare and communication systems), Wayren (digital communication platforms), or Krakul (internet of things for defence).

The Estonia cybersecurity sector also has enterprises that work on AI-enabled solutions. Chief among them are Cybernetica (encryption technologies), CybExer Technologies (cyber range technology and services), Guardtime (data security and blockchain technology applications) and others. Many of them are collaborative partners of the Foundation CR14 as well as Estonia-based NATO’s Cooperative Cyber Defence Centre of Excellence (CCDCOE)—important players in the defence AI development ecosystem.

In addition, two well-established Estonian companies are coordinating major pan-European consortia financed within the framework of the European Defence Fund:

  • iMUGS

Launched in 2020 and led by Milrem Robotics, iMUGS (Integrated Modular Unmanned Ground System) strives to develop “a modular and scalable architecture for hybrid manned-unmanned systems in order to address a large range of missions and to enable easy update or modification of assets and functionalities within the system (aerial and ground platforms, command, control and communication equipment, sensors, payloads and algorithms)” (European Commission 2020). It heavily builds on the TheMIS UGV platform and works on the premise that such platforms are supposed to be part of a broader system of systems where AI in C2 as well as AI-enabled autonomy will be major pillars of the new capabilities.

  • EUROGARD

Launched in 2022 and led by Baltic Workboats, this project is less ambitious in its scope than iMUGS and has set out to “build a vessel capable of a range of different autonomous operations in coastal areas” (European Commission 2022). In this case, Estonian leadership of the consortium draws heavily on the expertise in developing autonomous maritime surface platforms for civilian applications that has already seen through deployment of a research vessel capable of autonomous operations at sea in collecting various data (ERR 2023a).

Cybersecurity is viewed as a pivotal element of future secure, safe, and trustworthy AI systems, allowing Estonian companies to leverage their competitive advantages in larger projects led by their allies. As an example, Estonia’s Cybernetica signed a contract with the US Office of Naval Research to develop cryptographically secure AI. According to the company, “the PAI-MACHINE project will optimise algorithms for collaborative AI applications so that allies can share work together for the common good without having to share their confidential data in full” (Cybernetica 2023).

The opportunities related to the European Defence Fund are also pursued in this field. In the 2023 cycle of proposals, Foundation CR14 is leading AIDA (Artificial Intelligence Deployable Agent), a consortium of 28 organisations from 15 nations. The consortium aims to develop a set of Technology Readiness Level 7 software agent prototypes that rely on AI algorithms for their operation in the cyber incident management cycle. The project is expected to improve protection of the EU’s critical infrastructure (e.g. satellite communications), alleviate human resource availability problems in cybersecurity, and enhance military mobility through securing operations of autonomous vehicles and aircraft. The decision of the European Defence Fund is anticipated in 2024 (E-mail communication, 1 December 2023).

3 Organising Defence AI

Estonia’s approach to organising defence AI reflects its overall organisation for defence innovation. The system rests on the “bottom up” initiatives from within the defence organisation as well as stimulating and facilitating the innovation ecosystem outside the government structures—in the civilian universities and enterprises. It also focuses on enhancing relations with foreign partners. This ensures that ideas for development and experimentation are relevant to national and international end-users, while also staying connected to wider trends in the transatlantic and European arenas. At the same time, however, this approach lacks overall strategic coherence, continuity in projects, and momentum in harnessing such disruptive technologies as AI.

3.1 A Broad Alliance for Defence AI…

Estonia’s defence policy has always placed strong emphasis on the so-called “broad-based national defence,” whereby all national stakeholders—public and private—ought to have a role in ensuring the nation’s defence and provide capabilities for the common efforts (Estonian MoD 2010, 2011). When it comes to defence technology and innovation, this whole-of-society and whole-of-government approach in practice often means that the MoD spends portions of the defence budget for the projects of civilian research establishments or companies that have potential or clearly identified military applications.

The AI-related efforts at the national level—just as digitalisation of governance and cybersecurity—are coordinated by the Ministry of Economic Affairs and Communication. The MoD, in turn, has consolidated numerous activities within a single department of defence innovation, even though other departments such as those overseeing NATO and the EU affairs or national defence information systems, contribute to its mission from their respective roles. The department is responsible for formulating R&D and defence industrial policies. It has recently established a position for coordinating all AI-related matters and, eventually, developing the Estonian defence AI policy.

On a working level, however, the MoD must draw upon the expertise and contribution of state agencies and other entities outside its formal structure but within its area of governance. One such agency is the Estonian Centre for Defence Investment, where management of most procurement programmes was centralised in 2015. The centre prepares technical requirements of equipment acquisition and has specialised managers for various categories of equipment. AI will inevitably become part of these requirements across a range of capability projects, thus necessitating a certain level of competence and the capacity to incorporate impartial external advice from the national and allied S&T base.

Moreover, the MoD area of governance includes organisations that have great flexibility in combining various sources of funding, incorporating national and foreign public and private sector partners into common projects, and providing hubs for testing various practical solutions. Foundation CR14 is key to the Estonian efforts to advance cybersecurity and cyber defence solutions, with AI rapidly becoming an important part of this effort. CR14 operates cyber ranges that are used by various stakeholders (including NATO CCDCOE) to conduct exercises and test new solutions (Foundation CR14 Undated). These cyber ranges also help to generate datasets that can be used by the data owners (e.g., nations that provide “blue teams” at Exercise Locked Shields (NATO CCDCOE Undated) to train AI algorithms in cyber defence. The foundation’s CEO has also been appointed to represent Estonia in NATO’s Data and Artificial Intelligence Review Board (DARB).

The MoD also provides support for non-governmental entities and programmes initiated by other stakeholders. One such programme is Cyber North, a cybersecurity AI accelerator, launched by Start-up Wise Guys, a major Estonian technology start-up accelerator, in cooperation with the Estonian Defence and Aerospace Industry Association in 2019 (Enterprise Estonia 2018). Since then, it has become an accelerator within NATO’s DIANA network (ERR 2022). This not only brings the Estonian entrepreneurial expertise in start-up creation and growth to bear on developing AI-powered solutions for defence, but also enhances Estonia’s profile as a multi-spoke hub and integrator of such innovation benefiting the entire Alliance.

3.2 …Meets Defence Forces Not Yet AI Ready

While the MoD appears reasonably well organised and prepared to pursue defence AI-related ambitions, the EDF has a poorer organisational preparedness. For a start, there is no centralised innovation management function at the top-level EDF headquarters, and the office of the Chief of Defence no longer has a position of chief scientific or innovation adviser to provide advice on what should be included in the top-level defence planning and capability development guidance to the services and commands.

If any substantive consideration is given to the EDTs, and AI in particular, it should be included through the long-term planning processes managed by the defence planning department of the HQ (J5). However, as demonstrated by the failure to stand up an autonomy programme initiated to provide centralised coordination from the EDF HQ for this particular area of technology, it is all too easy to abandon organisational measures if they are not regarded as a strategic priority and do not have a powerful high-level advocate in the organisational hierarchy (Interview, 20 December 2022). Interest in and initiatives to advance innovative solutions that might include AI thus basically reside at the level of individual units and communities of practice (e.g. military intelligence, signals and communications, or electronic warfare).

The EDF, however, sought to centralise its interaction with the civilian S&T community and defence industry via the Applied Research Centre of the NDA. This has proven useful in providing a central hinge between external innovators and end-users within the EDF (Jermalavičius and Hurt 2021). However, it has not been able to build the capacity that is quite central for the military’s ability to define the need for innovative solutions—operational analysis/operational research (OA/OR). It is also pushed into too many directions, such as securing funding from the EU sources through consortia, supporting the educational mission of the academy, and addressing the requirements for research in social sciences, thus diluting its ability to focus on the application of the EDTs, including AI, in defence (Interview, 9 May 2023). It is also too far removed from the centralised planning processes at the EDF headquarters, making it less impactful than it should be in future-proofing long-term defence development plans (Interview, 21 December 2022).

Existing challenges notwithstanding, long-term defence planning may undergo some important changes soon. The ongoing project to redesign planning methodology will open some important opportunities to consider EDTs in the future capability mix. For example, it will identify points in time when various external (e.g. defence industry) or internal (e.g. the NDA) stakeholders will have a possibility to engage defence planners in a discussion on technological aspects of the capabilities. Services would be provided some funding to experiment with the technologies that may become part of their capability mix in the future, in the expectation that their analysis and lessons will shape their inputs into the defence plans (Interview, 14 July 2023).

Additionally, some of the more recent new force structure elements need to pursue new technological solutions within a very dynamic threat environment due to their core mission. Estonia’s Cyber Command, established in 2018, is one example. Its mission is to conduct operations in cyberspace, including information operations, making it one of the domain-specific combatant commands alongside land, air, and naval forces. It also provides support to other EDF services and commands as well as agencies in the MoD’s responsibility when it comes to information technology, infrastructure, and services, thus also making it akin to a C3I agency or combat service support organisation (EDF Undated). Uniquely among the EDF services and somewhat contrary to efforts to centralise EDF’s R&D coordination at the NDA, it is also tasked with R&D work in its domain, thus linking it with external partners in the wider defence innovation ecosystem.

By virtue of this triple role, but particularly due to its exposure to the domain where AI applications are emerging very fast, it thus potentially becomes a major gateway for AI diffusion into the military organisation, especially given the synergies and overlaps between cyber operations, ISTAR, electronic warfare, information operations, and C3I management. This, however, will be contingent on greater acceptance of its role and mission across the EDF, which is not always there. Furthermore, its success will also rest on how well it will be able to combine its various tasks, as currently its cyber and information operations, for example, are quite poorly integrated (Interview, 30 November 2023).

This structural arrangement is, to some extent, mirrored by the Estonian Defence League (EDL)—a paramilitary organisation for territorial defence manned by volunteers. Its Cyber Defence Unit was stood up in 2010 and provides a platform for the volunteer members of the EDL from all walks of life to exchange their knowledge, train together and act when required (Estonian MoD 2010). The unit extends its remit into information operations as well, and it would be natural to expect that, with the proliferation of AI tools in the civilian cyber and strategic communication sectors, members of this unit will bring ideas and solutions from these sectors to the military environment.

The discipline of long-term defence planning seems, however, to have been put on the back burner recently. This will most likely also affect its interplay with defence innovation. Currently, Estonian defence development is driven by the fundamental assumption that Russia—a primary and existential threat to Estonia—will reconstitute most of its lost capabilities by the end of this decade if not sooner, necessitating a steep and rapid increase in Estonia’s national defence capabilities. The country is now basically trying to compress a 10-year cycle into 4 years or even less to arm and prepare itself as much as possible ahead of what it regards as a quickly approaching point of a direct military confrontation with Russia (Toom Kooli Strategic Talks, 15 November 2023).

In this mindset, there is little space left for considering long-term implications of various EDTs—including in terms of interoperability with allies. Only innovative solutions that are mature enough, readily available and in use by key allies—and therefore can be quickly introduced into the capability mix or adapted to military uses from their civilian origins without negative impact on interoperability—draw attention of the planners and military leadership.

4 Funding Defence AI

For years, Estonia has been one of the few NATO Allies spending more than 2% of its Gross Domestic Product (GDP) on defence. Russia’s full-scale war against Europe has shifted the terms of the spending debate in the Alliance, whereby Tallinn is advocating this level as a floor rather than a ceiling. While Estonia’s defence budget of 2021 was €749M (2.2% of GDP), it grew to €1.1bn (2.73% of GDP) in 2023 and is expected to reach €1.3bn (3.2% of GDP) in 2024 (Riigikogu 2023a).

Such growth, however, does not deliver a very large pool of funding in absolute terms or create too much additional space for funding R&D/R&T. The needs for capability development and stockpile build-up are considerable and, along with the need to compensate for the impact of inflation on personnel pay, crowd out more long-term requirements. According to senior military officials, the overriding principle in budget allocation currently is whether a programme or project is likely to deliver tangible results (i.e. fielded capabilities) within the next four to 5 years (Briefing, 28 September 2023).

As the defence organisation essentially resides in the emergency mode, spending on long-term technology and innovation pursuits might sound, to a large degree, as a luxury item. As of 2021 (i.e., before the war in Ukraine), Estonia was spending just €5.1M on R&D and only €1.1M on R&T (using EDA’s definitions) (EDA 2022). However, it seems that the wave of fresh funding for defence is spilling over into the R&D investments, despite the long-standing scepticism and the pressure of current events. According to the MoD figures, its R&D investments were €5.6M in 2022 and €7.8M in 2023. They are projected to more than double in 2024 and reach €12.1M (E-mail communication, 20 November 2023), which can be explained by the need to co-finance involvement in the EU-funded projects. Still, this is only about half of the EDA’s benchmark of 2% of the defence budget to be spent on R&D/R&T.

Given current budgeting methods, no data is available to identify specific spending priorities such as defence AI, for example. This makes it very difficult to gauge the MoD’s funding levels for it. Nonetheless, some defence industry support grants of the MoD go to the projects the title of which explicitly or implicitly suggest the inclusion of at least some elements of defence AI (Estonian MoD 2023c):

  • In 2021, a project “Prototype of Unit of an automated mini-drone-station with ‘search & find’ AI detection model software.”

  • In 2022, development of an Intelligent Decision Support System (IDSS) and Mixed Reality Situational Awareness System.

  • In 2023, a project “Tactical Data Exchange Platform with Armoured Troops Awareness System Integration.”

These are usually small grants, ranging between €72,000 and €200,000, but they are an important element of stimulating the industry’s interest in defence markets, especially among young enterprises and start-ups. In contrast, Estonia’s civilian AI sector receives far larger sums from the private financial sector or external sources. The OECD AI Policy Observatory estimated that Estonia’s overall AI start-up ecosystem received a cumulative investment inflow worth USD437M as of 2023, which has almost tripled in size since 2021 (OECD Undated). Although this dwarfs anything that the MoD can offer to finance defence AI development, the benefits of such investments in the broader AI sector are bound to seep into defence-related projects at some point.

Multinational sources of funding like the European Defence Fund (and one of its precursors, EDIDP) as well as Horizons Europe (Horizons 2020 in the previous cycle) have emerged as pivotal in advancing AI-related developments in Estonia. The latter is still something that defence research and innovation projects are not eligible for (those are addressed via the European Defence Agency’s instruments) but given the dual-use nature of AI and the reliance of the Estonian defence innovation on the civilian S&T base, this restriction eventually becomes trivial.

The European Defence Fund, on the other hand, has direct relevance and significance to the Estonian enterprises involved in the defence sector. Estonian enterprises have been particularly successful in tapping into this source, with 16 international projects that involve companies from Estonia receiving an overall total of almost half a billion euros of funding. Six of those projects also receive Estonian MoD co-funding (Estonian MoD 2023a). Again, it is hard to isolate, without additional research, what share is devoted to AI-related technologies. Some of the MoD co-funded projects (e.g. Cyber and Information Warfare Toolbox (EUCINF), Collaborative Combat for Land Forces (LATACC), Naval Collaborate Surveillance (E-NASCOS)), however, are bound to include AI one way or another and may even source this technology from Estonia, judging from the participating companies such as Milrem Robotics, Cybernetica, Cafa Tech or Criffin that seek to incorporate AI into their products and services.

If and how these projects will deliver tangible capability gains for the EDF, however, is an open question. For now, the military is fully satisfied with the arrangement that minimally taxes the defence budget and shifts the financial and technological risks to industry and the EU, while drawing upon the military end-user’s knowledge and feedback. It is possible, though, that EU or NATO-funded development projects will eventually become a stimulant of interest and demand from the EDF when formulating their capability requirements, thus accelerating the adoption of technology that those projects seek to advance.

5 Deploying and Operating Defence AI

In a country without a formal defence procurement policy in place, it is difficult to assess whether AI-enabled capabilities are prioritised in any way—this would require analysis of technical requirements of each procurement on a case-by-case basis. However, as an informal matter, the Estonian MoD and its procurement agency adhere to the principle that past support for development projects does not warrant procurement of the outcomes. Rather tenders evaluate all offers and select the best match with the requirements while remaining agnostic about the source of technology (as long as it is not China, Russia, or other similar originators) or whether it has domestic roots and has drawn upon investments by the Estonian taxpayers (Jermalavičius and Hurt 2021).

Despite this, officials insist that most projects supported with MoD grants within the defence industrial policy successfully secure sales with the domestic customer (Roundtable, 2 September 2022). This is partly the result of the military’s familiarity with technology built in a close consultation process between end-users and project teams, thus ensuring alignment of the results with military customer needs. For example, the defence forces gained useful experience in operating robotic systems as part of field tests and on missions. The THeMIS UGV was operated by the Estonian contingent in French-led Operation Barkhane in Mali, where three EDF platoons used the platform for patrols and transportation of supplies. This deployment provided the producer with valuable insights into the performance of the platform under harsh conditions (e.g., desert terrain, hot temperature) and enabled the military to better understand the added value of such platforms and the challenges in operating them (Milrem Robotics 2020).

Assessing the actual deployment of AI-enabled capabilities, however, remains complicated as they often constitute the most sensitive part of the package, especially in such domains as ISR or cyber, thus making them subject to high levels of classification. It can only be inferred that procuring ISR UAVs from Estonian producer Threod Systems for tactical and operational level combat service support functions could potentially feature in-built AI elements—if not yet currently then definitely in the future. Likewise, with the ongoing implementation of the in-house projects, KOLT and TOORU, are opening possibilities for deploying new AI tools to assist the processes in these systems.

However, the major ongoing shift in defence procurements will be a game changer for future AI-enabled capabilities. During the first two decades or so since restoring the independence, the EDF had to rely mostly on donated equipment or purchases of second-hand systems, with some notable exceptions such as long-range air surveillance radars or short-range air defence and guided anti-tank missile systems. But over the last few years, we are witnessing a surge in acquiring brand-new equipment and weapon systems with cutting-edge technology (Gosselin-Malo 2023). Many of these systems like

  • Blue Spear (5G SSM) land-to-sea cruise missile system

  • Blocker PM16 naval mines

  • M142 HIMARS multiple rocket launchers

  • K9 Thunder 155 mm self-propelled howitzers

  • Iris-T SLM medium range air/missile defence system

  • Long-range loitering munitions (LMs)

will already include AI or provide possibilities for incorporating it through upgrades or connectivity with AI-assisted battle management systems. Participation of the Estonian national representatives in the multinational end-user clubs and their interaction with the suppliers will play a crucial role in avoiding costly problems of integrating AI add-ons in the future. At the same time, the fact that some of these procurements are already undertaken jointly with foreign partners—such as Blue Spear and K9 Thunder with Finland and Iris-T with Latvia—opens opportunities for collaborative technology adoption pathways as envisaged by the emergent defence AI policy principle articulated by the MoD.

Herein lies a risk that the Estonian end-users will just not make sufficient effort to fully and comprehensively understand the AI technology packaged in the acquired cutting-edge systems, as in some cases it will primarily be examined, understood, and vouched for by the lead partners in joint procurement projects. This may not lead to the outright adoption of a black-box solution that would contravene Estonia’s position which holds that operators need to understand built-in AI. Rather it might create a grey-box situation, whereby EDF operators will probably have some necessary understanding, but will depend upon the explanations of irregular AI behaviour from their foreign military or industrial partners.

6 Training for Defence AI

Doctrinally, the EDF is not well prepared to advance force-wide learning outside the whole-of-alliance framework, as there is no formal national military doctrine in place and the entire system often directly relies on Allied Joint Publications (AJP) and Allied Tactical Publications (ATP) (Interview, 21 December 2022; Interview, 14 July 2023). Without the national process to think through and develop conceptual ideas as well as to establish them as part of the Estonian doctrine, the EDF becomes highly dependent on the progress of NATO in incorporating AI into its AJP/ATP family of documents. This, on the other hand, may serve as an opportunity, given that a small country alone can hardly address the challenge of drawing an overarching doctrine for joint multidomain operations, for example.

Many of the preconditions for effective learning about AI, however, could be laid in the professional military education (PME) system, with the NDA and the Baltic Defence College playing a pivotal role. For the time being, however, AI is addressed as a fairly minor subject within larger technology and innovation modules and courses, even though air and naval officer training provides some more extensive exposure to this particular technology. Likewise, the training of non-commissioned officers (NCO) for the land component—by far the most dominant service in the EDF—does not provide any significant skillset in managing AI as part of the military capabilities and tactics, techniques, and procedures (TTPs) (Interview, 21 December 2022).

As the Estonian defence system relies heavily on the mobilisation reserve, conscript and reserve (re-fresher) training programmes naturally constitute the core of the training activities undertaken by the EDF. These programmes undergo continuous modifications and adaptation, depending on, among other factors, when the defence acquisition programmes and projects deliver new weapons systems and equipment or their upgrades. In most cases, conscripts—and the units manned by conscripts—must demonstrate their ability and be certified to use these systems before being transferred to active reserve after 8–11 months of service, while the EDF’s professional core needs to gain deep understanding of those systems well ahead of their actual fielding. This gives a few years head-start before any major influx of AI technology takes place—including potentially together with the joint procurement partners (e.g. Finland and Latvia), given that pooled training and logistics are often touted as key benefits of collaborative procurement projects.

As discussed, the policy thinking suggests that AI-enabled systems should not be used if the operators do not possess sufficient understanding of how the AI behaves in the respective systems. Therefore, readiness of individual and collective training system to deliver upon this principle becomes critical. Military leaders acknowledge that, to be a competent end-user of sophisticated weapon systems, many of which will be drawing upon AI, the EDF will require ever growing number of soldiers-technicians—with the backgrounds and skill sets related to STEM (Science, Technology, Engineering, Mathematics) disciplines—in addition to the usual soldiers-tacticians (Interview, 21 December 2022). This, in turn, shifts the onus onto the national higher education and vocational training system, whose graduates then go through the basic and advanced military training programmes. With STEM disciplines lacking popularity, the MoD and EDF are now working on co-operative arrangements with various civilian schools to give them a boost through scholarships and other incentives, in the expectation that future demand for technical skills—including those relevant to AI—will be met by sufficient supply from the national education system (Defence Resources Agency 2023).

The EDF is not arriving at the threshold of the AI era being deprived of talents. Using the ProgeTiiger programme framework and tools aimed to enhance technology literacy starting as early as kindergarten (Education Estonia 2021), the world-class Estonian education system (Schleicher 2019) is already dedicating attention to such subjects as coding and robotics from an early age. The EDF can tap into this trend through conscription—which also serves as one of the recruitment tools for full-time military employment—and, as the roots of KOLT demonstrate, is even capable of encouraging, capturing, and scaling up the innovations that arise from this talent pool. However, it is a different subject, whether it is capable of motivating and retaining the best and brightest in the fierce competition for talents in a small labour market where private enterprises in the high-tech sector offer much higher salaries and benefits.

The conscription and reserve training system is not a perfect solution, as military service is mandatory only for men and voluntary for women, thus depriving the EDF of easier access to a significant part of a demographically aging and shrinking talent pool. Notwithstanding the fact that younger generations might have natural instincts for handling high-tech systems, the current approach also fits relatively poorly with the world of complex technology as the effective mastering of new technologies requires longer periods than the current duration of conscript service or the amount of reserve refresher training. It also locks many of the full-time professionals into repetitive training cycles, giving them fewer opportunities to experiment, innovate, and develop new skills or tactics, techniques, and procedures (TTPs). Nonetheless, the current approach synchronises the EDF with the broader societal trends among which AI proliferation will inevitably be a major one.

7 Conclusions

Estonia’s gradual embrace of defence AI illustrates the challenges and opportunities of broader defence innovation in a country where strong entrepreneurial drive and the government’s willingness to facilitate and leverage technological progress meet the conservative and cautious military focused on quick build-up of capabilities required to counter an existential threat to the nation. The lack of a strong, sustained, and systematic demand pull for AI-enabled solutions from the military side is compensated, to some degree, by the enthusiastic, dynamic, and conceptually well-versed AI innovation ecosystem that has taken root in Estonia because of its strengths in cyber and robotic technologies. Russia’s war against Ukraine and emerging AI-enhanced capabilities in key allies such as the UK, the US, and France serve as additional motivating factors for the Estonian military to increasingly pay attention on how to apply AI to shape future warfare.

Despite its relative novelty, AI is not finding Estonia’s defence organisation writ-large just passively watching the incoming wave. Thanks to efforts by the MoD and industry, it is well integrated into NATO and EU fora where defence AI is a central theme. This also provides opportunities for Estonia to join leading countries in large EU-funded defence projects focusing on defence AI. The military has been open to experiments and provides feedback to developers, using the NDA as the main interlocutor. The Cyber Command and the communities of practice (intelligence, electronic warfare) are places where AI will be making major strides. The development of in-house projects is laying further ground for the future exploitation of AI. If successful, the reform of the long-term defence planning process will provide incentives for the military services, branches, and units to pay more attention to what longer-term future challenges might be and how to prepare for them.

Some preconditions for the effective adoption of defence AI, however, are clearly lacking. There is no vision in place for the EDF to become a data-centric organisation or how far-reaching its digitalisation will be. Without an overarching AI policy and a national military doctrine, a lot of AI-related (as well as generally technology-driven) innovation might come to lack the scale, coherence with the national defence objectives, and eventual impact. The overwhelming focus on short and medium-term capability requirements leaves long-term foresight side-lined and marginalised. Almost entirely absent OA/OR capacity and concept development and experimentation structures in the EDF make it difficult to systematically estimate the impact of emerging disruptive technologies on warfighting and cast the existing organisational practices in a critical light. This is further reinforced by superficial coverage of AI and related technologies at all levels of the PME system.

The above shortcomings will, sooner or later, be addressed. It is clear though that Estonia is not setting itself an ambition to be first adopter of AI in defence, but rather seeks to tread carefully and avoid costly mistakes, while remaining integrated with the efforts and networks of its allies. Estonia will likely continue leading in some niche areas, where national strengths and civilian inputs are evident, such as cybersecurity and cyber operations or software for autonomous platforms and sensors. However, for a small nation, being a solid, competent, and active team player will be more important in defence AI adoption than pursuing its own high-flying visions or revolutionary transformations.