RAA Liaison Letter 2024 - 2025 Edition

RAA Liaison Letter – 2024 / 2025 Edition 91 enhance Army’s ability to contribute to joint force kill chains, which in turn may reduce risk to the ADF’s more vulnerable naval and air strike platforms. Challenges to Adoption Notwithstanding its inherent battlefield value, the integration of 4IR technologies into the Australian Army’s fires capability will pose numerous challenges. These include integration of the technology within existing fires and targeting capabilities, implementation of appropriate control measures on autonomous weapons and kill chains, and the risk of overdependence. Given the likely difficulties in resolving these matters, 4IR technologies should not be regarded as a panacea for all of the military challenges that the Army may face in the 21st century. Rather, new technologies offer the potential to augment conventional fires and targeting capabilities. Integration. The introduction of completely autonomous technologies into military applications is a recent phenomenon, making the risk–reward ratio yet to be determined. Because of this, an arms race is currently being waged between prominent military powers for superiority in 4IR-enabled weapons and software. 39 Global powers such as China and Russia favour fully autonomous systems. 40 Unbounded by human control, such systems have the potential to support the development of the most superior weapons. By contrast, the US and its Western allies generally favour semi- autonomous systems, where human control is retained over every engagement. 41 This preference is driven largely by concerns regarding the potential indiscriminate effects of automated weapon systems. While bound by such ethical concerns, many Western nations are nevertheless concerned that they may fall behind in the emerging global arms race While bound by such ethical concerns, many Western nations are nevertheless concerned that they may fall behind in the emerging global arms race. As a result, they remain reticent to join international weapons agreements that require a ‘human in the loop’. 42 The climate of competition that exists around the acquisition and use of autonomous technologies will make it difficult for Australia to determine how far it should automate its own fires and targeting capabilities. Decisions will become even more challenging as newer and more potent 4IR-enabled applications are discovered. Ultimately Australia must balance its ethical obligations with the need to field a fighting advantage. Control Measures . Divesting aspects of battlefield decision-making to an algorithm introduces the risk that an autonomous program might deviate from acceptable rules, such as the laws of armed conflict. This risk is particularly relevant to LAWs as they use machine learning to designate certain objects as threats. 43 It is particularly challenging to generate input control measures that can reliably ensure that a target is not misidentified. This is because the dynamic battlefield conditions in which LAWs must discern valid targets are theoretically infinite. 44 The International Red Cross (ICRC) seeks to address this challenge by recommending that lawful autonomous engagements are confined to specific target types within defined collateral damage parameters. 45 While many Western nations agree with the ICRC’s recommendations, no international consensus exists. It is reasonable to expect that this lack of general agreement will cause delays and complications in weapons design and subsequent adoption by Western nations. Indeed, war ethics theoretician Ross Bellaby highlights this problem as one of the fundamental challenges obstructing the adoption of autonomous weapons. 46 The International Red Cross (ICRC) seeks to address this challenge by recommending that lawful autonomous engagements are confined to specific target types within defined collateral damage parameters In addition to the implementation of controls around battlefield decision-making, there is a need for control measures to be applied to automated targeting programs. For example, a key risk relates to the reliability of data on which automated targeting programs are based. Data is the ammunition of automated kill chains, and the quality of data the program ingests will determine how well it performs. In testing, machine-learning programs have been found to underperform in situations where intelligence data is scant and enemy misinformation designed to fool the algorithm is active. 47 This is why many Western nations