Finland has recognized artificial intelligence (AI) as a top priority for technological and economic development, setting ambitious policy goals in the civilian sector. For defence AI, policy objectives seem less ambitious, focusing on general guidelines. Defence AI has been piloted in administrative and support functions, permeating military capabilities only gradually.

The Finnish Defence Forces (FDF) tackles AI in the broader context of digitalisation. Its cross-cutting Digitalisation Programme envisions AI applications within the FDF’s internal development programs. A key element of the programme is educating staff in AI to identify use cases and fostering expertise for developing and procuring AI. National Defence University (NDU) incorporates AI in its curriculum at all levels.

AI is a priority research area for the FDF. Potential use cases have been identified in virtually all areas of defence, and various R&D projects are underway. Robotics and Autonomous Systems are one area where disruption is being fuelled by AI developments. Another potentially disruptive area of AI application is in dynamic management of the electromagnetic spectrum.

The FDF acknowledges the crucial importance of data for defence AI, in particular machine learning. The availability of training data for developing AI applications needs to be ensured by tackling legal and organizational barriers, improving data storage systems, and enabling data sharing without compromising integrity and appropriate protection of data.

Ethical and legal issues pertain to both autonomous systems and the gathering and use of data for machine learning. Vis-à-vis Lethal Autonomous Weapon Systems, Finland’s proposals have been pragmatic: rather than an outright ban, case-by-case scrutiny is proposed as a potential basis for regulation, backed by a conceptual framework.

International cooperation provides a force multiplier for R&D, with the EU and NATO being the main multilateral forums of cooperation and the U.S. and Sweden ranking among key bilateral partners. The ratification of Finland’s NATO membership in April 2023 is boosting possibilities for R&D cooperation.

Fielding of defence AI applications seems to be taking baby steps. Publicly available information paints a picture of AI being procured in military-off-the-shelf systems, even if the FDF’s R&D portfolio hints at the possibility of original applications. Finland’s acquisition of 64 F-35 fighters and the corresponding industrial cooperation with the U.S. will propel the FDF’s use of AI to a new era within the next decade.

Finland’s AI ecosystem builds upon a solid foundation of AI research and education, estimated to be the best in the Nordic countries. Paradoxically, shortage of talent is the bottleneck for Finnish companies in their effort to scale AI (Seehus et al. 2022). The defence sector and the civilian sector share a challenge of data management, partly exacerbated by legislative hurdles. One key challenge for the FDF is to make its organizational culture and procurement process more conducive to innovation and experimentation.

1 Thinking About Defence AI

At the highest policy level, guidelines on defence AI stem from the Government’s Defence Report of 2021. It recognizes digitalisation and AI as prerequisites for developing national defence and key drivers shaping the operational environment. The report underscores the link of AI with autonomous military systems but expects AI to have a major impact on all domains of defence, e.g. information processing, situational awareness, management of weapon systems and logistics (Finnish Government 2021).

The Defence Report outlines a broad objective for digitalisation.

to manage risks associated with emerging technologies, take advantage of opportunities, optimize activities, create new services, activities and knowledge, develop new abilities, and be involved in national decisions. A key objective is to develop abilities related to utilising information and knowledge and leading with knowledge, which can be reinforced with different artificial intelligence applications. Applications can be used to improve the basis for decision making, since information will be available faster and it will be more accurate. (Finnish Government 2021)

The Ministry of Defence (MoD) published Strategic Guidelines for Developing AI Solutions in 2020. Outlining a policy framework on the development and use of AI in the context of defence, the document sets five strategic guidelines:

  • Defence policies and programs on AI should be coherent, compatible and updated regularly;

  • Acquisition of AI research, development and maintenance of AI should be agile;

  • AI know-how should be continuously improved via staff training and recruitment;

  • Data should be made available and used with flexible techniques with up-to-date infrastructure;

  • Defence administration must ensure the legality and solid ethical foundation of its AI applications (MoD 2020).

1.1 Definitions of AI

No single definition of defence AI has been consolidated across the defence administration. Moreover, most policy documents do not recognize a “military AI” per se but provide a general definition of AI and then proceed to elaborate potential military applications. For instance, the influential MoD Strategic Guidelines for Developing AI Solutions (Ministry of Defence 2020) starts from a very broad definition: “AI enables machines to perform tasks for which human intelligence has previously been required.” The document notes that AI is best used for tasks where human intelligence falls short, for instance, when the amount of data or required processing speed is too high or if there is a “need for analysis independent of human factors.”

Even if an all-encompassing definition of defence AI is lacking, one important piece of conceptual thinking has informed most policy documents on defence AI since its publication: the conceptual framework for defence AI commissioned by the Prime Minister’s Office (Ailisto 2018). Situating AI in the broader context of digitalisation, it defined AI as software or technology that “enables machines, programs, systems, and services to function in a reasonable way as required by a given situation. A reasonable level of functioning requires the AI to be able to recognize different situations and environments and operate in accordance with how the situation evolves” (Ailisto 2018).

The conceptual framework recognized AI to be not one single technology, but a diverse group of methods, applications, and research areas. Consequently, the framework presents ten dimensions underlying defence AI (Ailisto et al. 2019):

  1. 1.

    Data analytics

  2. 2.

    Perception and situational awareness

  3. 3.

    Natural language and cognition

  4. 4.

    Human-machine interaction

  5. 5.

    Digital know-how in working life, problem-solving and computational creativity

  6. 6.

    Machine learning

  7. 7.

    System level and system effects

  8. 8.

    Computational environments, platforms, services, and ecosystems for AI

  9. 9.

    Robotics and machine automation: the physical dimension of AI

  10. 10.

    Ethics, morality, regulation, and legislation

1.2 Ethics and Regulation of AI

The Government’s Defence Report 2021 states that while “taking advantage of the opportunities provided by new technology, it is necessary to consider the related ethical challenges and legal limitations” (Finnish Government 2021). The MoD stresses the importance of complying with international legal and ethical obligations “in the construction and use of artificial intelligence,” highlighting the role of legality and ethics as one of the five Strategic Guidelines for developing AI (Ministry of Defence 2020).

The Strategic Guidelines feature a categorization of defence AI application areas helping to clarify certain ethical and legal issues. Figure 1 shows that most applications of defence AI are unproblematic from a legal and ethical viewpoint; that is, they are subject to the same stringent ethical and legal considerations as non-AI applications. AI-specific legal or ethical scruples mainly arise towards the bottom right corner of the image.

Fig. 1
A flow chart and a chart of Finnish categorization of A I application areas in defense. The flow chart has A I capabilities with administrative and operational capabilities, and personnel matters, C 2, support, and influence systems. The chart has yes, maybe, and no with yes for force and attack systems.

Finnish categorization of AI application areas in defence. Source: Ministry of Defence 2020

Full size image

Finland is fully committed to International Humanitarian Law (IHL) and an active proponent of its application to all aspects of warfare, including defence AI. Finnish defence experts note that the same rules of IHL must in principle apply to autonomous weapon systems as to conventional forms of warfare. Special care must be taken to ensure that new weapon systems and military AI really comply with IHL in all circumstances (Finland 2020). However, the Finnish defence administration “does not self-regulate more stringently than what is required by law” (Ministry of Defence 2020). The MoD stresses that national or international regulation must not prevent the development of ethically justified, necessary and appropriate AI based solutions (Ministry of Defence 2020). A specific challenge is posed by hostile actors that do not comply with international regulation. Hence, FDF develops know-how “preparing against a threat that does not adhere to international, commonly agreed restrictions on the use of autonomous weapons in the future battlefield” (Nieminen et al. 2022).

While cautious of the potential threats, the Finnish defence administration also recognizes the potential benefits of military AI and autonomous systems for improving IHL compliance of warfare. “Artificial intelligence can also be used to reduce human suffering” the MoD notes (Ministry of Defence 2020). And Finnish experts (Finland 2020) argue, if “AI enabled machine autonomy is applied to weapon systems with appropriate human involvement and by using ambitious ethical standards, it can also support humanitarian objectives, by allowing higher precision and distinction for military purposes.” An AI-enabled unmanned asset can get closer to its target than a manned unit would, therefore enabling far more precise situational awareness and targeting data. This may reduce collateral damage to civilians. The unmanned weapon can also abort its mission, if on-board AI infers that civilian collateral damage is imminent.

Finland participates in the United Nations Convention on Certain Conventional Weapons and its Group of Governmental Experts (CCW GGE) on Lethal Autonomous Weapon Systems (LAWS). Food for Thought papers contributed by Finland to the GGE and the EU feature potential principles for the regulation of defence AI. Grosso modo, Finland does not advocate an outright ban on all weapon systems possessing a degree of autonomy, nor is it in favour of complete non-regulation. Even if no consensus is reached on the definition of LAWS, regulation can still advance based on an agreed categorization or characterization of AI.

In Finland’s 2020 Food for Thought paper to the GGE a framework for the appropriate level of human involvement in LAWS was proposed. The five-phase framework outlines the required level of human involvement to ensure IHL compliance in operational use. The first phase is a rigorous and comprehensive weapons review in line with Article 36 of the first Additional Protocol to the Geneva Conventions. A second phase constitutes reviewing military doctrines and their operational and tactical implementation. The third phase reviews mission planning, and sets pre-defined boundaries for the operation of LAWS. The fourth phase deals with launch—to be decided by a human—and operation beyond the point-of-no-return, where human control is no longer present, but where advanced AI could still enable LAWS to analyse information and adapt its conduct. For instance, observing it has surpassed the boundaries preset for its operation, the LAWS could adapt or abort the mission. The fifth and final phase of the review regards monitoring and ending of the mission.

A contribution to GGE in 2021 by Finland further elaborates a practical approach by identifying clear boundaries for the application of IHL vis-à-vis LAWS. The idea is that by stating the obvious, one can better delineate the non-obvious. Obvious cases violating IHL would include, e.g., a system (currently existing only in science fiction) that would be completely autonomous, operating beyond any human involvement (Finland 2021). Conversely, many applications of military AI are clearly as unproblematic as traditional non-AI systems, for instance most dual-use technologies, solutions making use of AI as supporting elements of a weapon system controlled by humans etc. (Ministry of Defence 2020). The grey area, then, situated between the obvious cases, is where contextual assessment case by case is always needed, as no universal rules apply.

1.3 FDF Strategies and Programs

The FDF AI Roadmap of 2018 was an early attempt at an internal guiding document for applying AI in defence, informing FDF decision-making. The document did not set precise long-term objectives in an uncertain field, but rather proposed an iterative, evolutive and agile approach: technical design should not be too refined or robust at too early a stage since that could lead to unwanted path-dependency. Existing AI technologies and products were to be applied more extensively, without fixed long-term commitment to suppliers, technologies, or products.

While the Roadmap is a classified document, some of its key areas have been publicly highlighted (Heiskanen 2018): situational awareness and support for decision making; improved foresight; accelerating operational tempo; real-time sensor data fusion and analysis in service of situational awareness; establishing and communicating situational awareness for cross-sectoral cooperation between authorities; and applying AI for training and real-time simulation.

The FDF Research and Development StrategyFootnote 1 (2019) emphasizes AI as one of its priority R&D areas along with cognition and autonomous military systems. The Strategy underlines the need to assess the potential of AI across a wide spectrum of areas, ranging from logistics to ISTAR, from decision-making and C2 to management of big data. Consequently, also the NDU’s Department of Military Technology ranks autonomy, robotics, AI, and machine learning as one of their five main research areas for 2022–2026 (Nieminen et al. 2022). Moreover, the FDF identifies a few critical technologies intertwined with the development of AI: human and machine cognition, man-machine teaming, remote and autonomous systems, cognitive spectrum management, C2 and ISTAR, as well as positioning, navigation, and timing (Kosola 2022). Moreover, sensor and data fusion is a crucial area of application for AI. The basis for leadership and situational awareness is a networked data system, which combines information produced via sensor fusion with an AI that analyses it and provides solution proposals. FDF researchers point out that “(t)he impact of modern ground troops is based on data analysed by AI from a wide selection of sensor sources” (Tiilikka et al. 2021).

The FDF Digitalisation Programme (2021, 2022) is the overarching internal document for setting the pace and principles for AI development and deployment across the defence system. Digitalisation is defined as a cross-cutting functionality to be implemented via all the FDF sectoral development programs. The document encompasses plans for educating and training personnel on AI, creating and nourishing a digitalisation ecosystem, focusing on data, piloting a prototyping workshop activity, and establishing a process for gathering ideas as well as managing risks. Moreover, development processes need to be rendered more agile while promoting an organizational culture that encourages sharing and innovation (Karsikas 2022).

The Digitalisation Programme identifies processes and measures for harnessing digitalisation as a driver of change, enhancing the understanding of FDF staff on the possibilities of digitalisation for developing military capabilities. Concrete development projects will be based on selected use cases, with process owners in each service or branch leading their respective projects. The Digitalisation Programme highlights the central role of data as a prerequisite for making use of AI.

1.4 Data: Key Enabler, Key Challenge

Data is pivotal for digitalisation and AI. The MoD notes that “the amount of data has increased exponentially, which means that more efficient methods are needed to deal with it” (Ministry of Defence 2020). Finland’s EU Presidency Food for Thought paper (Finland et al. 2019) recognized obstacles for exchanging data even within the armed forces of one country, not to mention for the international sharing of data. Data needs to be stored and classified in such a way that will enable flexible and appropriate use in applications and pave the way for increased cooperation between EU Member States and NATO Allies. Indeed, one of the strengths of the EU should be the striving for joint procurement of materiel, pooling and sharing of equipment—and exchange of experiences and data.

The FDF Digitalisation Programme highlights the central role of data for improving current capabilities as well as enabling new capabilities (Karsikas 2022). This requires, on the one hand, that relevant data be made available e.g., for training AI applications, and on the other hand, that infrastructure and computational models be capable of handling the increasingly voluminous data masses.

The availability of teaching data is a key challenge for machine learning. Supplier selection for data gathering, management and storage solutions may inadvertently grant the supplier an advantage that may be unfair or even unproductive. The procurer must be alert, protecting the data used for training AI, so that the procurer retains full authority for future development of the solutions, also ensuring integrity against cyber threats. Turnkey procurement of AI systems is a potential pitfall for the uneducated client (Hemminki et al. 2021). A potential remedy for this may be found using synthetic data.

Various obstacles may hinder the efficient use of data, for instance if data is protected in such a way that relevant stakeholders cannot access it, or if data is stuck in silos due to organizational barriers, incompatible formats, or sloppy structuring. A special hurdle is posed by Finnish data protection legislation, which prevents the authorities, among other things, from using data in a database for any other reason than the one it was collected for. Rigorous data anonymization may provide a partial workaround.

One unpublished FDF study highlighted the need for quality inputs: in the context of predictive maintenance of armoured vehicles, it was found that when the original data was too ambiguous, neither a human expert nor the AI managed to make use of it. The AI can refine and use unstructured and uncategorized data but can’t deal with fuzziness any better than humans do. Therefore, maintenance data systems should provide preset, unambiguous structures for inputting data. Moreover, for wider application of AI, the maintenance system should enable reliable search functions for data allowing for workflow automatization, as well as develop a systematic, iterative function able to complete missing terms or correct errors. On the other hand, based on recent milestones in the development of civilian AI, some predict that AI can even be trained on unstructured data in the very near future.

The FDF is preparing a new Data Concept aimed at supporting the planning and use of the defence system by fostering better and more cross-cutting availability of data. A broad objective is to enable multi-domain operations jointly with allies. The guiding principle is to use data to support military and political decision making, based on an improved situational awareness. Ideally, situational awareness data will be compiled jointly among all NATO members in all domains. This will enable effective joint operations, including enhanced joint fires with an “any sensor to any shooter” approach (Solante, Interview 2023).

Achieving such objectives requires that data not be confined in silos but be made accessible between services and subsystems. This implies a paradigm shift in data security thinking—from “need to know” towards “need to share.” The approach is Data-Centric Security (DCS): security controls are aimed at the data itself rather than at the information systems. The objective is to combine data protection and sharing in an unhindered manner (Solante, Interview 2023). The FDF strives to implement DCS efficiently also in order to harmonize with NATO’s data architecture, of which DCS is an integral component.

A shift of organizational culture may be needed also to resolve the challenge of data ownership. In developing AI solutions, the FDF should take care to retain usage rights for the data machine learning algorithms are trained on—otherwise the supplier may get an unfair advantage over competitors, or an unhealthy client-supplier relation may emerge. When the use of synthetic data is possible, problems with data ownership should not arise.

Traditional acquisition processes pose a challenge to AI procurement due to the evolutionary character of machine learning. If the AI application is based on continuous improvement, it may not make sense to acquire it at full capacity. Moreover, machine learning solutions are rarely final but constantly evolve throughout their life cycle. Traditional procurement may therefore made obsolete by a more agile, incremental, and iterative acquisition model. This would require the FDF to significantly increase its tolerance of experimentation (Hemminki et al. 2021).

1.5 Ideal Roles for Man & Machine

Finnish defence policy documents reflect a way of thinking whereby AI and automatization complement but don’t replace human abilities. Autonomous systems are classically seen as well suited for dull, dirty, or dangerous (3D) tasks (Hemminki et al. 2021). Automation can improve cost-effectiveness of military systems as well as reduce the cognitive burden of humans. In cases where the amount of data exceeds human processing capacity or where the situation requires superhuman reaction speeds, automated or AI systems can bring capabilities to a new level. Moreover, one area that has so far been fairly uncharted is how soldiers’ social and ethical performance is impacted by the implications of human-machine teaming (Aalto 2022).

In discussions related to manned-unmanned teaming, the notion begins to emerge that a “wingman” approach may not be the optimal way to use unmanned assets. If the unmanned platforms are focusing on supporting the manned aircraft, the mobility of the faster aircraft is limited to that of the slower ones. Stealth capabilities would be only as good as those of the least stealthy member of the swarm. Therefore, it might be more useful to grant the unmanned assets more independence. Drones could be sent in advance to the area of operation, loitering in search of potential targets or commanded to advance to trigger the enemy’s air defence, followed by other drones tasked with a jamming or target acquisition mission; the role of the manned aircraft would then be to launch a missile from stand-off distance. An analogy is a hunter and a pack of hounds. As long as the hounds are on a leash, the team is inefficient. With the hounds unleashed, they can locate the prey and chase it into the range of a rifle. The decision to use lethal force is made by the human.

2 Developing Defence AI

2.1 AI in FDF Research

The Finnish Defence Forces conduct research and development (R&D) to generate knowledge that supports decision-making and to create the technological basis and knowhow for building and maintaining military capabilities. As Finland is a small country with a specialized but limited defence industry, much of the defence materiel is procured off the shelf. Consequently, much of FDF R&D focuses on experimentation, testing and integration.

However, a new Defence Materiel Policy Strategy states that “it is essential for national security that Finnish companies have an adequate technological level of know-how of critical technologies. Especially in digitalisation, AI, analytics, and autonomy, the security of supply for national know-how is an area of growing importance” (Ministry of Defence 2023). Certain capabilities need to be developed nationally to ensure technological sovereignty.

A rough outline of the FDF’s R&D modus operandi is as follows. Research topics are always based on defence capability needs. Low-TRL defence research is funded via public calls, such as the annual MATINE funding that links the defence sector with academia and research institutes. The most successful projects can be upscaled through FDF funding for higher TRL projects. International cooperation is used as a force multiplier when applicable. Once development projects mature, fieldable defence materiel is procured through the FDF Development Programs.

MATINE, or the Scientific Advisory Board for Defence, is a special structure established to ensure an active link between the worlds of academic and civilian research and the defence community. It promotes research on national defence and security while also functioning as a network of over 300 scientists. MATINE gives university professors a window into defence matters while functioning as an unofficial multidisciplinary think tank for the FDF. MATINE’s research funding focuses on risky or low TRL projects, the best of which can be subsequently upscaled with FDF funding.

The Finnish Defence Forces’ Research Program is a spearhead of FDF R&D. The current program (2021–2025) features several projects exploiting the potential of AI. It benefits from a particular feature of Finnish society, the long tradition of general conscription: since most men have completed military training, R&D procured from Finnish companies is inherently carried out by people with a hands-on military understanding. Detailed research objectives are not public, but project titles give a hint to the extent AI permeates FDF’s current R&D: the Situational Awareness portfolio of the program includes such projects as AI for Detection and Classification of Radar Signals; AI as Situational Awareness Operator and Analytical Support; Producing Situational Awareness with a Drone; AI in Processing Big Data Masses; and Target Situational Awareness and Sensor Fusion. Another portfolio is entitled Human-Machine Teaming, featuring projects like Autonomous Systems for Surveillance and Engagement; and Technology and AI-Powered Development of Operations.

International cooperation is a force multiplier for R&D. Much of the FDF’s multilateral R&D cooperation makes use of the technical expertise of the European Defence Agency (EDA) and the funding leverage of the European Defence Fund (EDF) and its predecessors. NATO is becoming increasingly important: Finland is already active in NATO’s Science and Technology Organization, participating in more than 70 activities, many of which focus on AI. Finland joined NATO’s Defence Innovation Accelerator for the North Atlantic (DIANA) in April 2023 and the NATO Innovation Fund (NIF) in May. DIANA helps identify and develop dual-use technologies with defence potential, while NIF provides funding for bringing innovation into the market. Spearhead Finnish accelerators and test centres for DIANA specialise in quantum technologies and 6G connectivity.

2.2 Potential Uses of AI

The FDF has identified numerous use cases for AI across all levels of the defence system, ranging from support functions to lethal force. The following is an attempt to summarize, based on publicly available sources, what types of use cases the FDF has identified as potential building blocks for AI-enabled capabilities.Footnote 2

AI could assist in forming troops by optimizing the assignment of tasks and missions, by providing a personal assistant to conscripts or monitoring the performance of soldiers or groups. Moreover, conscription could even be revolutionized by the introduction of virtual elements whereby a select portion of the training could be tailor-made and executed remotely. AI can support leadership and decision-making by compiling and analysing situational awareness data (Kallinen 2022), by formulating proposals and assessing the potential implications of decisions, by drafting orders and instructions as well as by synchronizing and monitoring execution. Moreover, AI could be used to simulate alternative decisions across a few scenarios far exceeding human processing capacity:

  • Intelligence

The constantly growing computational capacity of sensor systems, along with predictive analysis, enable a more and more complete and up-to-date situational awareness. Image recognition software is constantly improving and has, in certain cases, already surpassed human capability. Best results can often be achieved by applying AI in combination with human judgement, making use of the virtues of each.

  • Electromagnetic and Cyber Domain

Ubiquitous digitalisation opens up new pathways for intelligence data gathering from the electromagnetic spectrum as well as from the internet. As demonstrated by the war in Ukraine, and discussed in Vitalyi Goncharuk’s chapter, even data from social media can quickly transform into target acquisition. The cyber domain is a well-established area for AI for both defensive and offensive applications. Information warfare may be heavily exacerbated by AI which enables the automatic monitoring and targeting of people with low resources—a threat scenario to prepare against (Kosola 2021c).

  • Logistics

Logistics is already being optimised via such AI-powered processes as predictive maintenance and the automatization of stock and transport management. Military medicine is improving via preventive precision medication, and new ways of continuous measuring and enhancing of human performance are also made possible by AI systems. Searching for wounded soldiers can be improved by drones, and their evacuation can be carried out by Unmanned Ground Vehicles (UGV) with increasingly autonomous capabilities.

  • Force Protection and Engagement

More military-specific applications of AI include force protection and engagement. Protection can be enhanced by AI application via improved detection and identification of threats and automation of countermeasures on the one hand, and via improved techniques of concealment, decoys or misleading the enemy on the other. Engagement can be enhanced if the planning of kinetic force is improved via AI techniques; collateral damage could be reduced with AI-powered risk assessments; and AI can also enable advanced cyber or electronic warfare functionalities. Notably, AI applications can be used to enhance joint fires by exploiting targeting data from any sensor to any shooter (Solante, Interview 2023).

  • Acceleration of Decision-Making

AI can enhance every phase of the observe, orient, decide, act (OODA) loop. Decentralized AI applications, either in the physical domain in the form of robot swarms or as software agents operating in networks, can help to achieve a superior tempo of operations (Kosola 2020c).

All of the aforementioned themes are recognized by the FDF as potential areas applying AI. One broad theme is already emerging on the battlefield, partly enabled by the development of AI: unmanned systems with autonomous capabilities.

2.3 The Next Disruption: Robotics and Autonomous Systems

FDF Research Director Jyri Kosola estimates that the next disruption in warfare will be propelled mainly by unmanned systems and the combination of AI, digitalisation and data (Kosola 2020a, 2020b, 2020c). Analysing the possible ways in which robotic and autonomous systems could disrupt the battlefield, Kosola notes that unmanned sensor and weapon platforms can either act as a force multiplier for existing concepts or enable completely new doctrines of fighting. For instance, a traditional minefield based on guessing where the enemy might move and then deploying masses of stationary mines might be rendered obsolete by smart mines (Kosola 2021b). Areas and pathways to be denied could be decided only hours before the event and moving mines with targeting capacity free engineers from predicting enemy moves and making time-consuming installations. Moreover, fewer mines would suffice—and the blue force can pass through.

As the development of AI and sensor technologies enables machines to become increasingly aware of their own state and their environment, they become increasingly autonomous, requiring less external control. This evolution is expected to result in combat teams consisting of humans and machines in the 2030s. An appropriate division of labour retains humans in control of decision-making and monitoring, based on human capacity for situational awareness and contextual judgment. Correspondingly, the machine would play the implementing role, especially for 3D missions as well as situations requiring superhuman execution speed (Kosola 2020d).

How to use this disruption to gain operational advantage? Kosola reasons that this requires defining the man-machine division of labour already at the planning stage of operational concepts, thereby optimising the machine for its mission and conditions. For instance, an unmanned platform can be much smaller, since it doesn’t have to have space, life support or protection for a human. This enables improved mobility and resilience. The unmanned platform can also be more difficult to detect, enabling it to operate closer to its target.

A potential concept starts to emerge, one based on multiple small, inexpensive, and expendable platforms operating in a swarm-like fashion. Each unit may be much less capable than a large and expensive platform, but their large quantity more than compensates for the inferior quality. Expendability enables completely new concepts of operation. Kosola refers to mosaic warfare as opposed to monolithic capabilities. The swarm, enabled by AI-powered autonomous features, is stronger and more resilient than a corresponding monolithic capability (Kosola 2021a).

While these reflections do not necessarily reflect existing or even emerging FDF doctrine, they provide some insight into possible pathways into future capabilities. Many FDF research projects aim at creating a knowledge base and developing technological enablers that could be used as elements for various systems involving remote and autonomous platforms. A few such projects are introduced below.

2.3.1 Project iMUGS

Finland participated in the EU-funded R&D project Integrated Modular Unmanned Ground Systems (iMUGS) aimed at enhancing the autonomous features of unmanned systems and facilitating joint operation of machines and humans. The FDF’s main objective was developing knowhow and technologies as enablers for various autonomous solutions irrespective of supplier. Enablers include communications and navigation solutions capable of operating in Global Navigation Satellite System (GNSS) denied environments and sensors and algorithms enabling the platforms to cooperate. These also need to be resilient to cyber and electronic attacks as well as arctic conditions.

iMUGS achieved progress in autonomy, with UGVs capable of manoeuvring autonomously to pre-planned battle positions, choosing their trajectory, detecting obstacles, and navigating accordingly, also beyond line of sight. However, these feats were achieved only in a simplified environment; reliable and consistent autonomous navigation across a complex terrain such as thick forest remains a challenge. Swarming capabilities were demonstrated both with and without a link to a central controller, but the latter only in simulation. AI applications by Finnish companies included the development of swarming algorithms by Insta (machine learning for path planning and optimising swarm behaviour) and communications optimisation by Bittium (analysis of node data, smart routing, dynamic spectrum management).

2.3.2 Project Laykka

Several AI-related Finnish R&D projects are currently underway revolving around an experimental micro UGV platform named Laykka (Andersson 2022). It can be used for stealthy anti-tank missions, removing the human from the extremely dangerous task of destroying enemy battle tanks. It can also be used for intelligence missions, automatised patrolling and logistics tasks, medical evacuation, transport of ammunition, mobile communications relay station, as a UAV charging base or a loitering weapon (Hemminki et al. 2021). One particularly promising AI-based application is smart minefields: instead of mines being permanently placed at fixed locations, the mines can loiter for long times but also move around as necessary while the situation unfolds.

Numerous research projects led by the NDU are underway, each focusing on a particular field of narrow AI. In one project, AI functionalities are integrated into Laykka for field testing to verify simulation results from concept models run in a virtual environment (Nieminen et al. 2022). Another project develops military medicine solutions, especially casualty evacuation. A third project developed visual identification and classification of military vehicles using neural net algorithms with promising results (Hemminki et al. 2021).

2.3.3 Manned-Unmanned Teaming

The FDF collaborates with the German Bundeswehr and Airbus to develop capabilities of manned-unmanned teaming. A major milestone was achieved in 2022 with Europe’s first large-scale multi-domain flight demo held in Finland: two fighter jets, one helicopter and five unmanned drones teamed up to execute a mission under near operational conditions.

The fighter jets, helicopter and unmanned drones were connected via a meshed networking data link provided by Patria, allowing them to seamlessly interact, negotiate division of labour and switch control of unmanned platforms between several manned units. The remote carriers were commanded by humans aboard a fighter jet but executed much of their given mission autonomously. Drones with electromagnetic sensors detected enemy air defence positions. With visual confirmation provided by other drones, the fighter jet proceeded to eliminate the air defence.

2.3.4 New Concepts: Dynamic Electromagnetic Spectrum Management and AIMA

One area where AI may propel a breakthrough is dynamic electromagnetic spectrum management (EMSM). Equipping communications, radar and electronic warfare systems with cognitive capacities would allow the use of the electromagnetic spectrum in a dynamic way: A software-based transceiver could use AI to analyse the available spectrum, making optimal use of the spatially and temporally available bands and generating new waveforms depending on the frequencies available. With machine learning, the system could also extrapolate from previous experiences while adapting to new situations. The AI should learn to avoid interfering with civilian and blue force communications by dynamic use of frequency bands and could even execute simultaneous jamming or spoofing of enemy signals while providing blue force C2. The latter capabilities could be based, e.g., on signal modulation or polarization, combined with dynamic adjustment of output power. Moreover, machine learning may enable electronic warfare and intelligence systems to autonomously provide situational awareness of the spectrum and to identify anomalous signals and equipment.

While dynamic EMSM is an emerging area for R&D, the main hurdle for fully exploiting such systems might not be technological but regulatory. Current regulation of the use of the electromagnetic spectrum is fairly inflexible: the law simply divides the spectrum into frequency bands, which are then granted for or prohibited from use by defined operators. This does not provide for much spatial or temporal flexibility for dynamic AI applications.

Finland has launched international initiatives seeking to develop new capabilities applying AI-based autonomy combined with dynamic EMSM. The Permanent Structured Cooperation (PESCO) project “Arctic Command & Control Effector & Sensor System” (ACCESS) builds on the idea of developing an AI-based Multifunctional Aperture (AIMA) and transceiver capable of simultaneously providing mobile ad hoc network data link, localization and classification of enemy radio signals, blue force tracking, identification friend-or-foe, passive electronic surveillance, and even electronic warfare functions. AIMA should be both modular and scalable: bigger units in vehicles would be more powerful, while smaller units could be mounted on small UAVs working in swarms. AI-based swarming would make the system effective even if a single unit is not powerful. Smart emission control can help units get closer to the enemy, requiring less signal power (Kosola 2023). Combining AI, miniaturization and technological convergence, these projects may lead to novel capability concepts.

3 Organising Defence AI

On the national level, an ecosystem approach is applied both to the development of civilian AI and to the digitalisation of defence. Finland’s civilian AI ecosystem consists of over 400 startups (FAIA 2020), as well as more established software developers such as Reaktor and Futurice. The ecosystem is boosted by specialised initiatives, such as the Finnish Centre for Artificial Intelligence (FCAI) and Finland’s AI Accelerator (FAIA). FCAI is a community of experts promoting research, fostering linkages between the private and public sectors, while FAIA works to connect AI suppliers with organizations looking for solutions.

Defence specific AI development is fostered by the launch of Digital Defence Ecosystem (DDE) in 2022. With the aim of reinforcing competitiveness, cross-pollinating know-how and ideas, finding synergies and leveraging funding for key technology development, DDE strives to interconnect major defence industries, small and medium-sized companies and start-ups, academia, and end-users. DDE is industry-led with close connections to the FDF, whose role is to ensure the military relevance of project proposals, provide ideas and guidance for products that correspond to defence capability needs, and potentially contribute expertise and testing sites for projects.

Although briefly discussed as an organizational option, the FDF does not have a specialized AI agency but a matrix organization promoting digitalisation and AI. The FDF’s matrix organization is tasked with implementing FDF’s Digitalisation Program, whereby AI is to be mainstreamed into defence capability development programs. The main objectives of FDF digitalisation are capacity building for digitalisation and exploiting the value-added of digitalisation in the development, maintenance, and use of military capabilities. With top-down guidelines and interoperability requirements drawn from the programme, each service and each development programme come up with their own AI applications (Karsikas 2022).

4 Funding Defence AI

Funding volumes in the civilian sector are of interest, since that’s where most of AI development takes place currently, potentially spawning defence capability development via dual-use technologies. The Finnish Government launched a €200M investment package in 2018, financing AI innovations, development of know-how for AI technologies as well as enhancing public sector efficiency through AI applications. FCAI runs on a budget estimated at €250M for its flagship period 2019–2026. Its core budget is also frequently augmented by project-based funding from e.g. the Academy of Finland.

For defence specific AI funding, we need to turn to defence industries and the FDF. The FDF hasn’t published exact figures for AI, but its annual R&D budget is about €50M. AI being highlighted as a strategic priority area, this gives us some idea of the order of magnitude. Leveraging of national R&D investment is also being sought through international cooperation, increasingly via the EDF. An additional source of AI funding is embedded in FDF Development Programs other than research, which should contain mainstreamed AI applications along the principles of the Digitalisation Program.

5 Fielding and Operating Defence AI

Specific AI capabilities of the FDF have scarcely been publicly discussed: information on currently operational equipment is publicly available, but technical details on battle management systems, sensor systems etc. are mostly limited to information published by manufacturers.

The Finnish Army notes that AI is featured in dozens of applications within their operational systems. AI is most prominently used in areas such as support of planning, processing of geodata, data fusion, virtual assistants and other support functions, expert systems, simulations and wargaming. Other areas of application include machine vision and image recognition, predictive analytics, resource allocation, reporting, and various elements pertaining to unmanned and autonomous systems. The Army engages in specialized research to ensure that internationally developed concepts and equipment can be adjusted to conform to the Finnish conditions and particular requirements (Lampinen and Tahkokallio 2022). Predictive maintenance is one known area of AI application, with very high accuracy achieved in AI-based fault data analysis of armoured vehicles.

In the Navy, battle management systems feature some degree of AI in processing sensor information. For instance, Hamina class fast attack ships and Hämeenmaa class mine vessels are equipped with Atlas Elektronik’s ANCS combat system. Fire-and-forget missiles and torpedoes have a degree of AI for navigation and friend-foe identification (IFF). Rauma and Hamina missile boats are equipped with a tailored version of Saab’s RBS15 with inertia and GPS navigation. Hamina and Hämeenmaa class vessels also have the ITO 2004 air defence system equipped with Umkhonto missiles, processing sensor data for target acquisition and applying missiles with on-board inertial navigation and infrared seekers after launch. The Navy’s coastal mine hunter vessels of the Katanpää class are equipped with Kongsberg HUGIN and REMUS unmanned underwater vehicles (UUV).

The Finnish Navy is about to enter a new era with the ongoing Squadron 2020 project. The project involves the acquisition of four modern corvettes that will eventually replace seven current vessels to be decommissioned. Construction of the corvettes should be finished by 2026, with half of the €1.2bn budget spent on sensors, weapon systems and integrated C2.

The Finnish Air Force does not disclose details of how it applies AI in its systems, with the exception of certain projects related to logistics and predictive maintenance of Hornet F/A-18 s. Two of these projects are a failure prediction system based on machine learning and a Fatigue Life Analysis neural network model.

The failure prediction project was based on machine learning analysed data from the fighter jet’s equipment. An adequate level of accuracy proved to require extensive amounts of data, and therefore such systems are applicable only for equipment used frequently and yielding ample data. The HN F/A-18 s fulfil these criteria, with a single flight yielding millions of data points. Since the project results have been promising, such analysis tools will potentially be a significant aid to maintenance decision-making for future platforms.

In another project a neural network model was created by Patria simulating the structural stresses of the F/A-18 s using recorded flight data. The model was taught against direct physical strain gauge measurements from two aircraft, with the aim of reaching an adequate level of accuracy in predicting fatigue life for the rest of the fleet. First, a computational model of the HN F/A-18 was created, superimposed with an aerodynamic model, enabling the simulation of specific in-flight stresses. The model predicted the most critical points aircraft structure. Based on these predictions, onboard equipment measured physical stresses. The neural network AI was then taught to produce stresses from the physical measurements, flight data and engineering parameters. Similar systems may be applied in the next generation fighter aircrafts since the results were found to be both useful and cost-efficient.

The Finnish Air Forces are currently undergoing a generational shift with the replacement of Hornet fighters by 64 Lockheed Martin F-35A multi-role fighters. The largest public acquisition in Finnish history at €10bn, the F-35 s will be operational in Finland from 2025 onwards. The weaponry with which the F-35 s will be equipped involves AMRAAM, Sidewinder, SDB I/II, JDAM-family weaponry, JSM and JASSM-ER. Optimized during the procurement, the weapons package will be adapted to Finland’s operating environment and latest system upgrades (Ministry of Defence 2021). Moreover, the acquisition includes a voluminous industrial cooperation package—a potential force multiplier for FDF R&D including defence AI.

6 Training for Defence AI

Though a small country, Finland punches above its weight in AI: it produced the second most AI patents in Europe per capita in 2003–2017 (Ailisto et al. 2019). One strength of the ecosystem is cross-fertilization with neighbouring technology areas such as signal processing, electronics, edge computing and 6G, which have been successfully combined with AI. A prominent example is the cognitive sensor fusion development at Tampere University.

Finnish universities offer a broad array of AI related education. Master-level education and corresponding research is featured covering all the major subfields of AI: data analytics, perception and situational awareness, human-machine interaction, machine learning, problem solving and computational creativity, platforms, and robotics. Applied research in private companies focuses on data analytics, robotics, and perception (Ailisto et al. 2019).

The FDF recognizes the need for boosting education to apply AI across all areas of defence. This pertains to the whole spectrum of AI elements, ranging from perception and situational awareness to data analytics, cognition, computational creativity, and machine learning, from system effects and ecosystems to machine automation, human-machine teaming, ethics and regulation. Training and education for AI are also at the core of the FDF Digitalisation Programme. In-house expertise is being reinforced by supplementary education as well as new recruitments, coupled with procurement of external expertise provided by academia and industries.

The NDU is strengthening the role of AI in its curricula with elements of AI already present at all levels. AI also often features in theses in such topics as operational analysis and planning, battle management, internet of things, Big Data and machine learning and AI applications for specific weapon systems. Conversely, the use of AI and digitalisation to enhance FDF training processes on various level are being explored, with applications ranging from improved selection processes of defence staff, digitalising conscript training and enhancing simulators used for weapon systems training. In particular, the KESI simulator used for leader training applies AI for troop behaviour simulation and decision-making support functions (Rautio 2022).

7 Conclusion

Finland has ambitious national goals for AI, and there are ample policies guiding the development of military AI. Digitalisation of defence is underway with emphasis on support processes aiming and cost-effectiveness. Baby steps are now being taken in the direction of digitalising core military capabilities, but most public projects remain at the level of R&D.

The FDF acknowledges the central role of data for digitalisation in general and AI applications in particular. A new Data Concept seeks to enhance gathering and storing data, improve availability and enable more flexible utilization. Silos between data systems and organisational branches should be surpassed. Achieving such holistic data use would imply a real paradigm change. Finland’s NATO membership provides impetus for this development, with the Data Concept streamlined to fully harmonise with standards and procedures of the Alliance.

The defence administration has put effort into addressing ethics and regulation of AI. Their analytical framework indicates that most military AI is applied other systems than kinetic engagement and is therefore unproblematic. Scruples arise in the area of lethal autonomous weapon systems, and for this Finland proposes a conceptual framework outlining a pragmatic level of human involvement. IHL always applies both for man and machine, with the human commanding the troops bearing ultimate responsibility. In certain cases, IHL compliance can be improved by AI, for instance by enhancing situational awareness and the precision of targeting, or by enabling better decision-making through reducing the cognitive workload of the human.

AI will play a key role in enabling potential disruptions on the battlefield. One such disruption is already looming in robotics and autonomous systems, a key R&D topic in Finland. Autonomous systems could enable completely new ways of fighting, provided that an optimal division of labour between man and machine is inherently embedded in the concepts of operation.

Dynamic management of the electromagnetic spectrum is seen as an area that could potentially be revolutionized by AI. Communications or electronic warfare transceivers may soon feature spectrum situational awareness, autonomously choosing frequency bands and generating tailored waveforms. AI and machine learning could spot spatial and temporal margin of manoeuvre in the spectrum inoffensively to civilian frequencies, achieving blue force C2 without enemy interference, all the while intercepting or jamming enemy communications.

Indeed, Finland is advancing a project combining the disruptive elements of autonomous swarming and dynamic EMSM. Finland’s AI Multifunctional Aperture (AIMA) initiative and the corresponding PESCO project ACCESS strive to develop a system capable of simultaneously providing ad hoc mobile networking, electromagnetic spectrum situational awareness and electronic attack capabilities with a swarm of inexpensive, multifunctional units. A modular, scalable, and multifunctional swarm system could allow for previously unseen tactical concepts.

Despite a few innovative projects, Finland’s defence AI seems to be somewhat lagging behind the ambitious national AI policies, with the defence administration taking a cautious and very gradual approach. AI development is initially focusing on administrative and support capabilities as well as low to mid TRL research. However, Finland’s NATO membership, its F-35 acquisition and corresponding industrial R&D cooperation with the U.S. provide a boost to AI development. Cumulative developments may eventually bring about a real transformation of defence.