Implementing AI in defence is challenging because the high cost of failure places most defence applications within the high risk use case band. You can’t just deploy off-the-shelf AI solutions without thoroughly understanding the specific operational circumstances and adapting the AI accordingly. Here are my insights based on a decade of research with human-machine teams.
Things to consider before deploying AI in defence
Begin with a comprehensive understanding of the operational environment, including relevant procedures, human operator interactions, training requirements and performance thresholds under various conditions. A team of human factors specialists, psychologists and operations analysts will be able to help gather the fine details for analysis.
Assess whether AI can genuinely enhance efficacy and efficiency in the given context. Try not to become to fixated on one solution, there are loads of models to try out with a small group in a research setting. Assess things like ease of use, ease of integration with existing processes, team dynamics and of course performance.
Important requirements for defence AI
Military operators cannot afford to waste resources, effort and time finding workarounds for AI that has been poorly developed. These are the minimum requirements for your AI.
Performance: AI must deliver high accuracy and reliability, appropriate transparency, and explainability.
Robustness: The system should perform consistently across a range of real-world factors and be resilient to adversarial activity.
Manageability: AI solutions must support edge-capable updates and upgrades, often essential in defence scenarios.
Learning Capabilities: Given the limited data for rare and novel events, the AI must be capable of one-shot or low-shot learning.
Human-Centric Design: Ensure AI systems are designed with human operators in mind, facilitating ease of use and integration.
Context Specific: AI should be tailored and tested for the specific operational context in which it will be deployed.
System Integration: The AI must be designed with broader system integration in mind, ensuring compatibility and coherence with existing systems.
Thorough Testing: AI solutions need to be rigorously tested throughout the development lifecycle and continuously monitored in the operational environment.
Auditable: Ensure AI systems are auditable and comply with ethical and legal standards.
Engineering Principles: Design AI architectures in accordance with good systems engineering, safety, and human factors principles from the outset.
For further discussion on implementing AI in defence and how to navigate these challenges, please get in touch.