De Boisboissel, G. "Արհեստական բանականություն. կիրառման նոր ձևերը և ազդեցությունը զորքերի մարտական կառավարման վրա / Artificial intelligence: new uses and impacts on military command and control." Հայկական բանակ / Armenian Army, 2024, 36–70. https://doi.org/10.61760/18290108-ehb24.2-36.
Abstract:
General information and background on AI 1.1 The three battlefield revolutions The digitisation of the battlefield is a major revolution in combat, which needs to be assessed on a long-term scale as it will profoundly change military operating methods. First of all, it will mean that all the equipment deployed in the field will be interconnected with a tactical bubble that enables secure data exchanges to reduce the fog of war. What is already true for many armoured vehicles* will be true in the future for the dismounted soldier himself, who will be carrying advanced technologies. Processing these data will optimise military action. Firstly, through the speed with which information is processed, enabling greater responsiveness, and secondly, through the consistency with which information is processed, enabling omni-surveillance of the battlefield. But this revolution of the digitisation of the battlefield is coupled with a second one, that of military robotics. Among its advantages, of course, we have the removal of the combatant from the danger zone, a high-risk area where we would rather risk a robot than a human life (cave reconnaissance, mine clearance). If their energy autonomy can be guaranteed, these machines can also remain omnipresent in the field, where humans are subject to fatigue and climatic constraints (surveillance), or particularly in the 3rd dimension (flying over areas). Embedded technologies also enable them to be more responsive and more precise than human beings when carrying out a task. The enormous advantage of these last two qualities is immediately apparent if countermeasures need to be triggered to face a sudden threat, or if a favourable opportunity arises. More specifically, the use of robotic resources extends a military unit’s range of action beyond its traditional limits, traditionally established by its firing range. Carrying functional modules on robotic platforms will extend the unit’s information-gathering capabilities (remote cameras, sensors for CBRN threats detection) or its identification capabilities (algorithmic image processing), thereby extending the limits of its information-gathering range. Nevertheless, the use of these robotic systems requires providing a high degree of autonomy in their movements in order to reduce the human cognitive load induced by their control. This autonomy will be a factor of power, whatever the environment in which these systems operate (land, air, sea, submarine, space or cyber) and a factor in levelling the asymmetry of military potential. Recent conflicts (the 44-Day War of the Nagorno Karabakh conflict, more recently Gaza, Yemen and the Red Sea, and above all the Russian-Ukrainian conflict) have marked a turning point in the way UAVs (Unmanned Aerial Vehicles) are perceived. They have gradually highlighted the inevitability of war between robots, that autonomy will amplify in the years to come, whether in high or low-intensity, or in symmetrical or asymmetrical conflicts. A third revolution, that goes hand in hand with the first two aforementioned, is to be dealt with by this article: Artificial Intelligence (AI). This is a veritable tool at the service of the military that will enable them to manage some of the complexity of tomorrow’s battlefield, and in particular the multiplication of operational data, interconnection of deployed equipment with remote support systems, at combined, joint, or even allied levels. Such an abundance of military data to process is accompanied by a cognitive overload that is too significant for the military leader, requiring automatic data processing. AI is a response because, with its computational capabilities, it will enable heterogeneous multi-source data to be processed, real-time analysis and rapid responses, allowing for advanced automation within systems, priority management, optimisation of available resources, etc1. In more practical terms, it can be divided into two main categories: a) decision support for the military commander when preparing or conducting a mission, b) management, coordination and interconnection of multifunctional robotic systems. 1.2 The different types of AI This article will not deal with the difference between the various types of AI. Indeed Artificial Intelligence is a term that encompasses two very different notions: symbolic AI first of all, which can be described as a top-down model, in the sense that it simulates or describes a formal representation of thought, whatever the substrate on which it is based2. It is a transcription of human decision-making mechanisms into algorithms, which thus execute a thought formalised by the designer, but which therefore cannot deviate from the original framework in which it was conceived. The solutions found by these systems are therefore logical, but do not deviate from the rules that have been set. Then the connectionist AI, or neural networks, based on a simplified model of the biological neuron and its links with the synapses send to the AI information to be processed. Such a neural network must be trained to perform a specific task or to acquire new skills, the performance of which can be improved with experience. This is known as machine learning. We are entering a new range of AI here, one that moves away from the manual writing of computer programmes. A connectionist AI can be discriminative and focus on classifying the data it analyses, or it can generate content, such as images, videos, music, texts or 3D models. Unlike symbolic AI, it raises the question of the trust that a military leader can place in such a system. 2. The benefits of AI for the military With data being set to be ubiquitous on the battlefield in the future, the opportunities offered by AI to process data from the military world are manifold. We will attempt here a general functional approach to its possible uses. 2.1 Mission preparation AI will help military leaders to make better decisions in increasingly complex tactical environments. It will be able to study several alternative solutions and propose decision options based on the analysis of multiple parameters. Our brains are not efficient at making decisions when more than five or seven factors are taken into account depending on the person and the context; beyond that limit they go into cognitive overload, which often translates into an emotional burden for the military leader. AI has no such limits and can take hundreds or even thousands of factors into account! Prior to operations, AI will thus help the leader to prepare the mission and plan the operations: by analysing the 3D terrain mapping (lines of sight, radio coverage, hydrography, soil survey, inhabited areas), by reading the history of enemy operating methods in the area, by taking into account the expected meteorology, etc. It will be a decision-making aid and will be able to propose a choice of route according to the weather conditions. It can be used as a decision-making aid, proposing an optimal itinerary based on these elements and the history of the area (mapping of IED hot spots), the light and shade for movements, checking accessible high points and listing possible areas for UAVs landing or for searches, etc. Above all, the AI’s computational capabilities will enable it to compare the military commander’s courses of action with the enemy’s possible courses of action, incorporating a host of possible non-compliant cases (complex enemy attacks such as drones swarms or combat helicopters, jamming effects, electromagnetic attacks, deception, etc.). It will enable a set of candidate solutions to be proposed to the commander, who will then be able to decide on the best course of action. 2.2 Mission conduct In the conduct of a mission, by capturing and analysing data from the battlefield in real time, AI will give military commanders a better understanding of the tactical situation. There are many ways of doing this, including tracking people (facial recognition) or vehicles (shape recognition), and detecting enemy attacks (source of sound or light flashes). This requires sensors to be autonomous in their data processing, independently of networks involving on-board computing capabilities (i. e., edge computing). 2.3 Detection / Prediction The proliferation of cameras integrated into camouflaged and abandoned sensors on the field, or mounted on drones or microsatellites at altitudes that allow them to monitor the entire battlefield, will partially lift the fog of war for those who control them. AI will enable the detection of aerial stealth targets, and the spotting and identification of objects on images or videos taken by these various types of equipment, using conventional, IR or thermal vision. Data produced by these cameras and sensors are legion in theatres of operation, but armies suffer from a chronic lack of human resources to analyse them. Detection and identification will therefore be based on AI-assisted remote surveillance systems. Before deployment and under supervision, it will also have to learn how filter out false alarms, such as the rustling of leaves in trees due to the wind or the falling of dead leaves, which must not trigger alerts on images used to detect enemy movements. For detected and identified threats, AI will be able to predict and calculate the trajectory of targets in real time, and suggest priorities to deal with the fastest (missiles or remotely operated munitions). It will determine likely modes of progression for enemy vehicles or armed groups. AI will also be able to optimise radio transmission and coverage capacities according to the constraints of the terrain (mountains, relief, weather) and the resources available (positioning of communication relay drones). 2.4 Collaborative Combat Provided data is effectively shared between equipment, AI could encourage the emergence of collaborative combat as it makes it possible to optimise the distribution and availability of critical resources and effectors (i.e., means having an effect on its environment: jammer, smoke bomb, grenade etc.). Collaboration can be seen here on three main levels: a) the inter-environment availability of resources and effectors available in a given air-land tactical zone, b) cooperation between combat units and robotic systems during combat phases, and c) the organisation of logistical support by anticipating supplies as close as possible to the units, depending on the conduct of the manoeuvre. The necessary fusion of various types of data and the capacity for geo-distributed processing nevertheless requires very strong connectivity between the sensors deployed to carry out collaborative combat and a permanent flow of data. This requires an unjammed tactical network, backed up by connections to low-earth orbit satellites. 2.5 Equipment customisation In the future, AI will make it possible to customise the equipment worn by the soldier, i.e., his weaponry or the objects he wears, such as the exoskeleton. The exoskeleton will be able to adapt to the individual’s specific movement characteristics: each soldier having his own gait, AI will be able to learn it and optimise the exoskeleton’s muscular support accordingly. AI will provide cognitive assistance to the soldier through cognitive interfaces that are easy to use and present contextualised and adaptive information with a mental representation tailored to each individual, based on natural interaction between the soldier and the interface. Given the influx of operational and terrain data, it will be necessary to determine beforehand those of the soldier and personalise them: for example, with an intelligent filter adapted to the individual’s hierarchical level (group leader, platoon leader, captain) and his military speciality. Here again, AI can play a role in this filtering. Finally, AI can offer instant language translation capabilities for soldiers in the midst of a foreign population, adapting to local dialects and accents. It can also offer a “Speech to Text” capability for transmitting digital orders or chatting, adapting to the language of each person and its potential distortion depending on the context (as with the effects of stress, or as for pilots at high altitude subject to pressure variations). 2.6 Predictive maintenance In the field, equipment is subject to severe constraints. For any military equipment or weapons system, AI will help to improve their Maintenance in Operational Condition through predictive maintenance. It will enable self-diagnosis of vehicles or equipment, with access to external databases for diagnostic assistance in the event of breakdowns, but above all on a preventive basis. To achieve this, integrated HUMS (Health & Usage Monitoring System) will enable equipment to observe its own operating status. 3. The indispensable contribution of AI to robotics 3.1 Navigation AI will gradually be integrated into mobile platforms, and more specifically into robotic systems that include some form of autonomy (UAV air/ UGV land/ USV and UUV sea). Primarily for navigation functions to avoid a teleoperator being constantly dedicated to piloting and having consequently his cognitive load being dedicated solely to this function. It will enable robots to adapt to spatial configuration and unknown environments whenever necessary. It will also enable trajectory adjustments to be made under time pressure, particularly when unexpected obstacles appear along the way. Finally, it will enable these platforms to dodge threats and to position their effectors quickly and reactively. AI will also make it possible to overcome jamming constraints. While this article is being written, in the context of the Russian-Ukrainian conflict, we are close to observing remotely operated munitions that will be automated in their last trajectory section to track and neutralise the target, without direct human control. This also raises the question of a prior validation of the system’s activation by the military commander, who is responsible under International Humanitarian Law. Remotely operated munitions are currently heavily jammed in the last few hundred metres, and AI target identification functions will soon be developed to ensure the success of strikes in heavily jammed environments. 3.2 Managing multiple robotic platforms Swarms of multi-function robots, which can be multi-environments too, represent the next step in the introduction of robotics into the battlefield. Swarms can effectively occupy aerial or land spaces, ensure saturation effects thanks to redundancy of action and their sheer number. Several robotic platforms will be able to be coordinated by a collective intelligence, which adapts to external events and enables a “group behaviour” capable of reconfiguring itself and reallocating tasks internally to achieve a common objective. Their move will adapt dynamically, reactively and rapidly to the spatial configuration and to unknown environments, depending on the collective resources available. This will have multiple advantages: piloting will be supervised by a single operator, requiring less cognitive effort, and the swarm will be assigned a mission whose various components will be carried out by each of the specialised platforms (observation/detection, neutralisation, jamming, etc.). AI will grant them with a global strategy in the action, defining the expected characteristics of the swarm (speed and 3D device positioning), and the coordination of effects (observation, jamming, neutralisation, etc.). However, these strategies require modelling that takes into account the potential attrition of resources, but also the best configuration to generate a strong psychological impact on the enemy. Digital simulation is the technological solution that will make it possible to test on a larger scale various options to configure swarms and their possible uses, and to select the most appropriate configurations3. It will allow for testing vast combinations on the basis of several parameters: the rules of engagement laid down by the operational situation, the principles of the Law of Armed Conflicts, but also the types of swarm formations, the automation capabilities, and so on. 3.3 AI creates innovative robotic behaviours AI will also be innovative for robotic systems to which the military commander has delegated the execution of certain tasks. It will enable them to adapt their behaviour according to criteria that are no longer the classic criteria of an operation mounted with human partners, but mounted solely with machines whose loss in the field is entirely acceptable. In this way, it is possible to conceive a use centred on a main effect, whatever the attrition of robotic resources. It could be noted that these robotics systems are expendable and therefore have to be low cost and considered as consumable munitions, which in itself is already a conceptual evolution in military thinking. This gives rise to a number of exclusively robotic for which new doctrines of use could be devised, with some freedom of manoeuvre entrusted to the AI. For example, missions to deceive the enemy to provide support for a manoeuvre carried out by ground units. This can be done by deliberately misleading the enemy as to the direction of the friendly manoeuvre, with robots moving in the area where the enemy’s attention is required, or by disrupting them with trajectories that appear erratic or even incoherent. All this combined with the advantages offered by land-based robotics, such as responsiveness and precision, the effects of submerging by sheer numbers, and the ability to remain in the area 24 hours a day, this, provided that the robots have sufficient energy autonomy or that they carry out norias between their launch base and the action zone. While respecting IHL, we can imagine some AI whose objective will be to constantly harass enemy units by creating a feeling of constant observation and stalking, with the effect of depriving the enemy of the feeling of security that is essential to avoid any psychological collapse in the long term. On a more offensive level, AI will make it possible to seize opportunities in military action, in particular with the use of lethal assets integrated into larger, multifunctional robotic systems. The example of remotely operated munitions is very significant here, as they can be the assets around which the AI will organise the manoeuvre to detect and neutralise potential targets. For example, AI will be a particular component of future air raids in hostile territory for trajectory optimisation, in day and night conditions, and also for the training and positioning of robotic carriers and their effectors according to the potential risks detected. Compliance with IHL is of paramount importance in the execution of these missions, but its application to AI requires a specific development that could be the subject of another article, given the complexity of the subject. 3.4 Delegating tasks to AI The military leader will be able to delegate tasks to systems with a certain degree of autonomy, enabling them to carry out their mission 24 hours a day in the field, which is impossible for human beings, and will have the capacity to be more reactive than humans and therefore better able to react to saturating threats. The example of robot swarms is particularly telling, because with the advent of these sets of multifunctional robots, the leader and his subordinates will no longer be able to operate each of them remotely. He will delegate to a collective intelligence the piloting of each of the robots in the swarm, reserving for himself the piloting and control of the whole entity. Furthermore, since command performance is linked to responsiveness, and since machines are more responsive than humans, AI will be better suited to immediately seize opportunities or react to threats, especially saturating ones and to attrition. However, this delegation of tasks to machines is a new concept for the military that raises the question of subsidiarity and the trust placed in these machines. We will come back to this in the command chapter. 4. AI to help the weak against the strong Military superiority often remains the prerogative of States benefitting from technological advances over their adversaries. But how can a State protect itself if its technologies are less developed than those of a Nation with great technological and industrial military capabilities? To answer this question, it appears that AI can be a factor in levelling asymmetry on the battlefield, by making appropriate use of the capabilities it offers. The creativity of AI can indeed offer innovative solutions to a military leader to counter the doctrines of employment of enemy military equipment. An AI trained in knowledge of friendly and enemy doctrines can develop, as an example, surprise strategies with the assets that a leader has at his disposal in a given tactical situation. We will consider here two types of assets: military equipment served by human assets and those served by robot assets. For the former, AI will be limited to being a decision aid. For the latter, the AI can manage the robot assets on its own, if this task has been delegated by the military leader. 4.1 AI as a decision-making aid for daring doctrines AI can revolutionize military tactics through daring reasoning that can surprise the adversary. However, this requires excellent knowledge of the tactical situation, provided by a comprehensive overall view of the battlefield, often called God’s eye, thanks to cameras embedded in microsatellites as well as tactical drones. This knowledge of a precise tactical situation becomes entirely possible through the interconnection of one’s own or allied observation systems, which ensures a centralized vision of friendly and enemy troops movements. This is currently the case of the digital platform used by Ukraine in the Ukrainian theater of war, which is used to centralize all images emitted by drones to make them available to their allies. The military genius will then be able to refer to it and propose daring manoeuvres focusing on the detected enemy’s weaknesses and based on the knowledge of its modes of action in order to constrain its posture. 4.2 AI as a manager of robot assets Technological developments will very soon enable the development of coordinated systems of different robots, with a relatively low acquisition cost compared to traditional military systems, thanks to the reuse of civilian robots facilitating a “low cost” effect. AI will allow them to ensure the autonomy necessary to carry out the tasks that the military leader delegates to them, while maintaining constant supervision. Armed forces will therefore be able to benefit from their use. The responsiveness of these systems, in constant flight in the sky or easily deployable from a platform or from a truck, will enable them to counter saturating threats in real time. As a result, a battle for the occupation of 3D space by robotized machines is taking shape, setting drones against drones, swarms against drones and swarms against swarms. 4.3 Digital deception by AI While camouflage remains a basic rule for protecting your units, the introduction of AI brings a new discipline that will have to be integrated into deployed or embarked combat units: digital deception. Digital deception is a new discipline whose aim is to prevent the enemy from collecting exploitable data on our forces, but also to deceive opposing AI, which will thus be disrupted in their process of analyzing captured images. In addition to classic concealment measures, notably to conceal our forces and equipment from the omni-surveillance from the sky (satellites, drones), deceiving the enemy will also involve camouflaging our military data captured on the battlefield. This will involve breaking down shapes (characteristic points) and electromagnetic signatures. Indeed, the capture of imagery intelligence (IMINT) is now increasingly delegated to remote systems, using cameras that most often only render a 2D image, i.e., with no relief effect, on which the precise detection of details remains complex. Mirrors, for example, can become a simple way of deceiving, which is very difficult for AI to detect. The latter usually relies on detected shapes. For connectionist AI, certain neurons will specialize in the recognition of specific shapes. Let’s take the example of the tank: an AI specialized in tank recognition will have learned specific shapes characteristic of a combat tank (turret, cannon, tracks, and so on). Adding whacky shapes that are incompatible with a tank structure will most likely confuse the enemy AI. It is no longer a question of breaking lines as in classic camouflage, but of adding patterns that will lower the AI’s statistical detection of a tank. For example, by adding wooden panels with painted window shutters to the tank’s superstructure, on all 4 sides and on the top of the tank. This may seem very incongruous, but the AI will be completely confused and will not be able to conclude that an enemy tank is present, as the risk of error is too high. Of course, the enemy will quickly become aware of this, but by the time he has performed a new training for his AI to get around the problem, the friendly forces will have time to modify the paintwork on the wooden panels, and replace the shutters with … other motifs such as slates! It may be objected that this type of superstructure paint will not be compatible with traditional camouflage, the aim of which is to deceive the human eye. Not necessarily, since the primary aim of visual camouflage is to break the classic shapes detectable to the human eye. The solutions outlined above are admittedly simplistic, but further study will enable us to reconcile these two constraints: breaking the lines, and adding structural elements unlikely to be seen on military equipment. Another example is the use of fake images, with deliberately modified dimensions. These can deceive an enemy AI into not making the correlation between an object’s size and its environment, because AI simply hasn’t learned perspective in the sense that we humans manage it with our binocular vision, which gives us an idea of distances in 3D and a perception of relief. In the same vein, a virtuous AI will have learned the basic constraints of international humanitarian law (IHL), implying the non-aggression of civilian populations during combat. This represents an ethical and technical challenge, which requires AI to respect these unbreakable rules. Consequently, simulating civilian personnel on the same tank (by adding mannequins placed on the vehicle), will once again disturb an ethical AI trained not to open fire on suspicion of firing on defenseless populations. Reasoning can be applied to electromagnetism. The difference is that it is difficult to reduce an electromagnetic signature, which indicates emission on the battlefield, and therefore active participation in the conflict. This makes it difficult to mask one emission with a different electromagnetic signature. 5. The impact of AI on command For the military leader, the issue raised by the introduction of AI on the battlefield lies in the constant need to retain his responsibility for decision-making. Nevertheless, the three revolutions outlined in the introduction will gradually call into question traditional military command, which until now has been reserved for Man, the only one capable of apprehending the context and the environment and synthesizing the information4. 5.1 From vertical to horizontal command structure Primarily, since the dawn of military history, the strength of the armed forces has always been their hierarchical command structure. Soldiers place their trust in their leader, who in turn is aware of the overall tactical situation and decides on the idea of manoeuvre. Subordinates carry out orders and trust their leader’s tactical analysis. However, with the digitization of the battlefield, data is now more widely available than ever before. What is known as the verticality of command is thus disrupted by the horizontal nature of the information disseminated to all. As a result, the leader will have to formulate his orders keeping in mind that his subordinates also share the same information, and may even have received and analyzed it before him. He must therefore take their opinions into account and co-construct his thinking with them, at the risk of otherwise cutting himself off from a host of advice from his staff deployed in the field as close to the action as possible. 5.2 Optimising the decision-making process duration At the same time, AI and its ability to process information in real time means that military decision-making duration can be shortened. In a conventional decision-making process, which is traditionally broken down into four steps (information acquisition, analysis, decision based on certain rules or constraints, then action or lack of action), immediate access to information means that the decision-making process is extremely shortened. Besides, the decision-making process itself can be delegated to a machine, whose computing capacity and reaction time are far superior to those of human beings. 5.3 Hybrid subsidiarity Finally, command is exercised through subsidiarity. This will involve two components: the traditional one with the soldier team member under his command, and the future one with AI-integrated systems to which he will delegate tasks to carry out. The leader will therefore have to proceed in two phases. The first phase consists in defining precisely the tasks he delegates to these systems, while ensuring that he controls the framework within which their actions are carried out. He defines the actions that have to be validated at his level, the others being subject to a regular report. He also controls the temporal and spatial framework in which these systems evolve. Once these systems are active, the second phase implies to command by reaction. This takes place at a more global level, as in the example of the swarms mentioned above. It can be both a) a command by reframing if the action carried out by these systems deviates from the spirit of the manoeuvre intended by the leader, or b) a command by veto that temporarily or definitively stops the triggering of actions considered critical. It should be noted that the ethical question raised here is that of the trust we can place in machines that are potentially more efficient than humans, but which are not moral agents in the sense that they will never be aware of the scope of their actions and decisions. They are simply algorithms that execute. In contrast, the soldier exercises discernment and free will. 5.4 Maintaining the demands of command However, whatever the assets at his disposal, the military leader must take responsibility for the military action he leads. This principle is both structuring and reassuring. Structuring, because it ensures the credibility of the command, which takes responsibility for its own actions despite the fog of war. Reassuring, because the leader retains the need for discernment before making any decisions, and avoids offloading his responsibility onto the behavior of the machines at his disposal. Moreover, the military leader takes decisions according to the context. He is the only one able to take into account the global situation of a military action, to see beyond the initial data of the mission he is leading, and beyond the data emitted on the battlefield. Besides, he is the only one who is aware of the moral implications of his actions – something, let us not forget, any machine would never possess. Nevertheless, he will have to train himself to avoid the new challenge of the significant reduction in the time allowed to make a decision, as outlined above. AI will certainly enable systems to react more quickly, but this advantage is the same for the enemy: “He who shoots first wins”, as Lieutenant-Colonel Rommel wrote in his memoirs. He will therefore have to train himself to discern and decide promptly, in order to maintain his superiority over his adversary, while restraining himself from the temptation of excessive confidence, or even fascination, in the AI’s performance. Consequently, AI, as a built-in tool in military systems that can allow for a degree of autonomy, must not cause the military leader to lose the possibility of regaining control over the machine. Here we quote Professor Dominique Lambert, who lists the conditions of supervision required to ensure that the human sense of the action is preserved5. According to him, human supervision must be: sufficient, which means that humans introduce into the management of the weapons system sufficient conditions (and not just a few necessary conditions) to ensure that ethical principles are preserved and that the rules of International Humanitarian Law and the rules of engagement are satisfied; meaningful, meaning that it is ultimately always a reference to the human sense that must guide the design, development and use of weapons systems […]; coherent, which means that at no point can the weapons system contradict what human authority has prescribed as the goal of action. In fact, it would be incoherent if a weapons system deployed to fulfill a certain mission began to behave in a way that is inconsistent with the prescribed aims. 5.5 Defining a national strategy for a sovereign military AI Given the risks inherent to this new technology, every country needs to set a strategy that respects the ethical constraints it has defined. We can cite here as an example the French Army which, in its AI Task Force report of September 2019, indicates the need to: rely on trusted, controlled and responsible AI; maintain the resilience and scalability of its systems; preserve national sovereignty; maintain freedom of action and interoperability with its allies. These general principles for a strategic policy for the implementation of a sovereign AI can be taken up or adapted by Armenia, partner country of France. The fact is that connectionist AI relies on the data that feeds its machine learning process, and then for the data processing to be carried out. But, if the data with which the AI of the civilian world is trained (Gemini, LLaMA, ChatGPT, etc.) is plethoric, it is because they have been generously and freely made available to GAMMA by their owners, without them even realizing it. The same cannot be said of military data. Under no circumstances should military data be distributed to the whole world on an open-access basis, and it must remain the priority of sovereign states. Military data is of vital importance! This means knowing how to retrieve and preserve it: it is a challenge to national sovereignty. As a result, these same states need to develop their own sovereign AI, which are the only ones to be authorized to use their military data. While the help of allied nations is invaluable in this respect, the use of military data by foreign AI will have to be agreed on between the countries concerned. It is also conceivable to start with neural networks developed by foreign nations and specialized in a given function, and then to enrich them by learning new additional classes which will remain the property of the sovereign country. 6. Conclusions As a logical consequence of the digitization of the battlefield, the interconnection of systems and the drastic increase in the amount of data to be processed, AI offers a host of opportunities that every nation must seize by adopting a development strategy that enables it to retain sovereignty over its own military data. In terms of command, AI is not just another technique. It requires every military leader who uses it to master its use, to seize the opportunities it offers, while understanding its risks. This implies having leaders capable of grasping the complexity, and not simply delegating its management to technical specialists. While AI military experts, engineers and technicians are obviously needed, so too are military personnel and officers trained in these techniques. To achieve this, the latter will be able to draw on simulation tools to help establish new doctrines of use for AI-built-in systems (detection, counter-threat, robotic systems of systems, swarms), as well as tactical situation exercises in which AI is used as a decision-support tool for manoeuvre preparation, or in conduct. We however must bear in mind that, while AI enables us to express a new type of military genius in the service of our forces, the enemy can also be inventive and changeable, with the sole aim of surprising us in order to win. References 1 See Gérard de Boisboissel. Déclinaisons et applications possibles de l’IA dans le domaine militaire. “Moroccan National Defence review”. First edition, Juin 2023. 2 See Dominique Lambert. Que penser de… ?: la robotique et l’intelligence artificielle. Fidélité/Lessius. Editions Jésuites, 2019, N 100. 3 See Thierry Berthier, Gérard de Boisboissel. Du drone au essaim de drones: une nécessaire modélisation des comportements au profit de la simulation. Conference CAID DGA 2023. 4 See Gérard de Boisboissel. Intelligence artificielle et commandement. “Défense et Sécurité Internationale”, Janvier-Février 2024, N 169. 5 See Dominique Lambert. Fondements éthiques d’un approche humainement signifiante du problème des SALAS. “Les enjeux de l’autonomie des systèmes d’armes létaux”, Pedone, 2022, P. 138. * The SICS (Système d’Information du Combat de SCORPION) is an on-board operational information system all French Army vehicles are gradually being equipped with.