Source: Stockholm International Peace Research Institute (SIPRI)

The Stockholm International Peace Research Institute (SIPRI) has published a detailed analysis on the critical issue of bias in military artificial intelligence systems and its profound implications for compliance with International Humanitarian Law (IHL). The publication argues that AI systems, particularly those used for intelligence, surveillance, reconnaissance (ISR), and targeting, are susceptible to various forms of bias that can compromise their ability to operate in accordance with the core principles of the laws of armed conflict. Bias can be introduced through skewed training data, flawed assumptions in algorithmic design, or the operational context in which the AI is deployed. For example, an AI system trained on data from one region may perform poorly and make erroneous classifications when used in a different demographic or environmental context.

The report meticulously examines how such biases can directly undermine the fundamental IHL principles of distinction, proportionality, and precaution. A biased AI targeting system might struggle to accurately distinguish between combatants and civilians, or it may miscalculate the potential for collateral damage, leading to disproportionate attacks. This creates significant legal and ethical challenges, as it raises questions about accountability when an autonomous or semi-autonomous system makes a decision that results in a violation of IHL. The authors stress the urgent need for states to develop and implement rigorous testing, validation, and verification (TV&V) protocols for military AI to detect and mitigate bias throughout the system's lifecycle. The publication concludes that without addressing the problem of bias, the use of AI in armed conflict could lead to a severe erosion of civilian protection and undermine the very foundations of international humanitarian law.