Machine Learning in Defense: Ethical and Legal Insights

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Oct 25 2023

In a recent publication in the journal Machine Learning and Knowledge Extraction, researchers conducted an extensive examination of the ethical and legal ramifications of the integration of machine learning (ML) into defense systems. They navigated the challenges, advantages, and disadvantages and presented an illustrative project while setting the stage for a profound contemplation of ML's implications within the defense realm.

Background

The advent of artificial intelligence (AI) and ML has ushered in a transformative era across multiple industrial sectors. Remarkably, ML applications, including virtual assistants, speech recognition, and text-to-speech technologies, have found a niche in the defense sector, significantly enhancing security. This sector, closely linked with Public Protection and Disaster Relief (PPDR) and Mission-Critical Services (MCSs), shares a common mission of ensuring public safety and safeguarding critical infrastructure. The collaborative integration of AI and ML emerges as a pivotal force in driving technological innovation within these domains, thus fortifying emergency response capabilities and bolstering resilience.

Methods

The current study drew upon the extensive military backgrounds and practical interactions with ML applications. The primary objective was to scrutinize the intrinsic issues associated with the utilization of ML in military contexts. The aim was to furnish insights that facilitate well-informed decision-making regarding the effective deployment of ML in this sector.

Data Collection: Data acquisition involved a dual approach, combining structured technical research with insights from real-world projects.

Case-Study Selection: For a comprehensive analysis, the authors strategically chose a pertinent case study, the SALAs project, encapsulating diverse ML applications within the defense sector. Additionally, this case study integrated findings from a survey that gauged attitudes toward the utilization of lethal autonomous weapon systems.

Problem Identification: The amassed data underwent a methodical examination, aiming to discern recurrent challenges that had emerged from the authors' practical experiences. To provide a comprehensive perspective, the authors classified these challenges into four distinct dimensions: technical, ethical, operational, and strategic.

Comparative Analysis: The authors conducted a comparative analysis of the identified challenges, with a specific focus on the military context. Although these challenges could theoretically apply to artificial intelligence in a broader context, the analysis delves into their implications within the military sphere. Drawing on their collective experience, the authors derived practical recommendations for the amelioration of challenges and the advancement of successful ML integration in defense.

Integration of ML in defense sectors

The application of ML in defense holds promise for enhancing security, decision-making, and efficiency. It parallels civilian applications, indicating a transformative shift in defense. Various defense systems have already incorporated ML-based technologies, including unmanned aerial vehicles (UAVs), to analyze sensor data to predict equipment failures, bolster cyberattack defenses, provide real-time situational awareness to military personnel, and identify potential threats. The integration of AI and ML in defense calls for robust legal frameworks and ethical scrutiny. Outdated regulations and a lack of consensus on ethical principles pose challenges.

Legal and ethical frameworks

Historically, emerging technologies have necessitated legal and political interventions, accounting for ethical, health, and social concerns. ML, with its potential for autonomous decision-making, follows this pattern. The European Commission (EC) took a significant step by issuing a "White Paper" in February 2020 in Brussels, a milestone in European AI regulation. Its primary aim was to cultivate a thriving AI ecosystem within the EU, outlining seven prerequisites for AI legislation, encompassing human oversight, technical robustness, privacy, transparency, non-discrimination, fairness, and accountability. To protect fundamental rights, a member country of the European Commission initiated a regulatory pilot project, the "sandbox," in June 2022. This project will shape the forthcoming European Regulation on AI over the next two years.

While ethical dilemmas arising from AI and ML have been explored in various sectors, the defense sector remains relatively unexamined, necessitating a comprehensive analysis. In the current context, adherence to International Humanitarian Law (IHL) is crucial to minimizing the humanitarian impact of armed conflicts. It requires human supervision of warfare to ensure proportionality in protecting non-combatants and mission success.

Military AI, despite its benefits in speed and precision, raises concerns due to its lack of human judgment and moral reasoning. It can lead to challenges in distinguishing between civilians and combatants or responding proportionately to aggression. Ethical considerations include accountability, preserving human dignity, and ensuring that autonomous algorithms do not undervalue human life. The international survey conducted by the Open Roboethics Institute (ORI) in 2015 highlighted the ethical disapproval of lethal autonomous weapon systems (SALAs), with 67 percent advocating for an international ban, emphasizing the prominence of ethical concerns over wartime strategies.

ML applications and challenges

The Advanced Targeting and Lethality Automated System (ATLAS) project aimed to equip US combat tanks with AI and ML capabilities for three times faster target identification and attack. It developed a learning algorithm to process sensor data, automatically detect and identify threats, and assign weapon orientation and elevation for an attack.

ATLAS focused on data collection, image processing, trigger control, technical support integration, and sensor deployment, including visible sensors, gyro mechanisms, and lasers for real-time data provision to the ML algorithm. ML holds great promise in military applications, but challenges persist. Key challenges include misidentification of assets, adversarial attacks, transparency for user trust, ethical aspects in decision-making, and data scarcity and values.

Conclusion

In summary, the impact of ML on defense is profound, bringing efficiency and cost savings. However, legal and ethical challenges arise due to autonomous decision-making. The current study explored the need for comprehensive frameworks while respecting sector confidentiality. To harness the potential of ML in defense, such as improved training systems, managing complexity, and adhering to legal and ethical standards is crucial.