February 2022 - Leveraging AI to Mitigate Civilian Harm

For example, the incident featured a misassociation: the pilot wrongly associated information regarding the threat with the location of civilians and friendly forces, leading the pilot to engage the wrong location and cause civilian and friendly casualties. [...] To better define the specific problems that AI applications should seek to solve, we performed a meta-analysis of the thousands of incidents we have examined, defining the specific mechanisms that led to civilian harm over the last decade and a half, including in Afghanistan, Iraq, Syria, Somalia, and Yemen. [...] In response to command emphasis on being effective and yet sparing civilians in Afghanistan, forces there actively sought to find solutions that presented fewer risks to civilians in their missions while preserving the success of the mission and the safety of the force. [...] This approach represents civilian harm through collateral damage due to the unnoticed presence of civilians in the area of fire, which is exacerbated by the tactic developed to mitigate friendly fire but not to consider the presence of civilians. [...] Since then, Ushahidi has been in continual development and has been used to track the evolution of a variety of regional events, including the disaster relief efforts following the earthquake in Haiti in January 2010, the Syrian revolution in 2011, and the earthquake in Nepal in 2015.57 As another illustrative example of the power of crowdsourcing, a recent paper in Armor: Mounted Maneuver Journal.