This paper outlines the requirements for Joint Global Command and Control (JADC2) based on embedded artificial intelligence/machine learning (AI/ML), explains how to leverage AI/ML systems in JADC2, the obstacles that need to be overcome, and points out the development path.
Establishing an "information foundation" is a prerequisite for applying artificial intelligence (AI) and machine learning (ML) to Joint All-Domain Command and Control (JADC2), requiring that the data within this foundation be tagged, secure, complete, comprehensive, and easily accessible. Building this information foundation requires the continuous organization and protection of all cross-domain, cross-service, and cross-echelon information needed by the military for command and control operations. This information serves as the input for AI and ML algorithms. Achieving JADC2's objectives depends on identifying the command and control requirements of the core military mission set and establishing software development plans that are achievable in both the near and long term.
This article outlines the requirements of Jointly Built-up Command and Control (JADC2) based on embedded artificial intelligence/machine learning (AI/ML), the obstacles that AI/ML needs to overcome, and points out future development directions. The article argues that JADC2 requires investment in talent and resources to break through today's labor-intensive command and control (C2) model. Improving the current planning process through autonomy and AI/ML is a valuable and realistic goal.
The necessity of JADC2 for supporting multi-domain operations
Modern warfare demands that commanders plan, command, and control forces across land, air, sea, space, cyber, and electromagnetic spectrum domains, requiring flexible and secure means to achieve cross-echelon, domain, organizational, and geospatial communication and data sharing. Future all-domain warfare will place even greater demands on the rapid acquisition and interpretation of massive amounts of information and rapid decision-making; these are all key elements of JADC2 capabilities.
Current systems and infrastructure used for military planning, scheduling, and monitoring are no longer suitable for modern multi-domain operations. Given the complexity, timeliness, and massive data requirements of multi-domain operational planning, military planners need to develop and utilize new tools, especially those based on AI/ML. Therefore, it is essential to first understand the capabilities, challenges, and application potential of these tools to determine the priority of their deployment.
Figure 1. Types of Machine Learning
The appeal of AI/ML
Recently, AI/ML systems have demonstrated capabilities beyond human reach in increasingly complex games. Coupled with a growing understanding of the demands of future high-level conflict warfare, AI/ML has become extremely attractive. As an AI/ML system, AlphaStar's success in the real-time strategy game *StarCraft* suggests that supervised learning and reinforcement learning may be applied to command and control at the tactical and operational levels in the future. However, transitioning these technologies from games to warfare still requires extensive research.
As artificial intelligence algorithms are developed for use in real-world, multi-domain, dynamic, large-scale, and complex rapid operations, it is necessary to select, evaluate, and monitor key metrics to measure their performance, effectiveness, and applicability. Key algorithm metrics will include the following:
efficiency:
Calculation time and memory required
reliability:
Does the algorithm produce valid results?
Optimality:
Does the algorithm provide the best result for a given objective?
Robustness:
Can the algorithm smoothly degrade under unexpected circumstances?
Explainability:
Can humans understand the reasons behind the results?
Certainty:
Does the algorithm run as expected?
To determine which AI/ML technology to adopt, the military must first understand the operational requirements these technologies can support (e.g., air superiority, air defense, refueling support, etc.). These operational requirements will then determine the command and control processes needed to achieve the mission, such as situational awareness and airspace conflict resolution. Understanding the limitations of AI/ML technologies is equally important, especially the difficulties these technologies encounter when reasoning under uncertain conditions.
Figure 2 AI/ML Relationship
Obstacles to be overcome in realizing military applications of AI/ML
Several obstacles exist in realizing the military applications of artificial intelligence/machine learning:
Obstacle 1: Differences between military culture and business culture
The military prioritizes information security (information should only be accessible to those who "need to know"), while the business world values open access to data (widely shared for application development and other economic benefits). Therefore, incorporating security considerations into military software development and IT operations (known as DevSecOps) is crucial to thwarting adversaries and dangerous actors attempting to undermine command and control through cyber means. Perhaps the biggest challenge is ensuring AI/ML algorithms are applicable to the real battlefield. The uncertainty, information asymmetry, and adversary actions of the real battlefield are drastically different from those in a game environment.
Obstacle 2: Data is inaccessible within the military.
The military needs unified data management policies and sufficiently advanced information technology to enable command and control personnel to access massive amounts of data, thereby supporting AI-assisted decision-making. In other words, an AI ecosystem supporting the collection, tagging, storage, protection, and sharing of data is essential. This ecosystem will rely on common data standards, explicitly defined permissions, integrity checks, and intrusion prevention. Cloud computing and data lakes will be key components. Cloud data lakes can be used for distributed computing, redundant storage, and connectivity across the enterprise. Given existing military policies, culture, permissions, budgets, and access methods, building such an environment to deliver massive amounts of data in a secure manner across domains and security levels will pose a challenge to JADC2.
Obstacle 3: The need to reorganize military operations centers and train personnel to manage them.
Increased machine-to-machine communication and the autonomy of command and control processes may lead to changes in the hardware and personnel configuration of operations centers, enabling human warfighters to focus on cognitive tasks such as assessing and refining potential course of action. The adoption of AI/ML technologies will create new roles and responsibilities. Warfighters will need to be trained to manage and operate AI ecosystems while also acting as data administrators, ensuring the quality and integrity of data captured and stored within those ecosystems. Furthermore, while planners and decision-makers are currently trained to think within a single domain, new responsibilities may emerge requiring people to think across multiple domains simultaneously.
Obstacle 4: The existence of military subculture
Due to subcultural differences among combat personnel, varying planning timelines, and different permission allocation schemes adopted to achieve different operational effects, integrating AI capabilities across the air, space, and cyber domains is challenging, even within a single service. Nevertheless, the need for comprehensive command and control is urgent and increasingly prevalent. Therefore, embedding AI applications in JADC2 must confront and overcome all of these obstacles—military culture, cybersecurity issues, algorithms used for problems with poor knowledge quality, data inaccessibility, operations center reorganization and training, and military subcultures.
Path to achieving efficient development
The report identifies effective human-machine collaboration in command and control as the objective and recommends timely progress towards JADC2 within the adversary's defense timeframe. First, the JADC2 operational concept should be further developed and its priorities established, followed by leveraging AI/ML to enhance capabilities in command and control processes. Simultaneously, conditions should be set for a data-driven AI ecosystem, migrating weapon systems and related data to a multi-domain data lake for those with authorized access, and applying zero-trust and other security principles to manage this data flexibly and securely. As AI software applications are developed, it is necessary to test these applications in operational testing environments, integrate them with C2 systems, then deploy limited capabilities to operations centers, and rapidly update the software applications based on user feedback. Analysts and technical experts can further explore operational concepts to facilitate human-machine collaboration, establish trustworthy AI, and improve the interpretability of algorithms.
Lacking real-world data to inform improvements to these warfare technologies, the military can leverage modeling, simulations, and exercises to generate training data for AI/ML algorithms (such as weapon-target pairing). Supervised or reinforcement learning algorithms can support this command and control function, but military algorithms must also account for real-world uncertainties, a major challenge for both humans and algorithms.
