Who Decides Who Dies? Artificial Intelligence in Armed Conflicts

In the context of the Israeli-Palestinian conflict, the use of advanced artificial intelligence (AI) technologies has marked a radical transformation of military operations. The “Lavender” system, developed by the Israel Defense Forces (IDF), represents one of the most controversial and discussed applications of AI in the modern battlefield.

A New Operational Paradigm

Based on deep learning algorithms and data mining, the Lavender system was designed to identify military targets – particularly members of Hamas – through automated analysis of a wide range of data, collected through mass surveillance. Along with the complementary software “Where is Daddy?”, it has enabled the identification of tens of thousands of potential targets, associating each with suspicious behavior or a recurring pattern.

The approach introduces a new paradigm compared to the past, when human evaluation was central to the targeting process. With AI, the procedure is structured in five phases:

1. Identification of the suspect through AI.
2. Association of the target with a residence.
3. Selection of the appropriate weapon (drones, precision munitions, or conventional bombs).
4. Authorization of collateral damage, with initial policies accepting up to 15 civilians killed for each member of Hamas.
5. Automated calculation of collateral damage, based on estimates via mobile data.

The Effectiveness and Limits of the System

The model has proven to be fast and efficient, but not without ethical and operational issues. For example, estimating collateral damage based solely on the number of cell phones present in a building has caused significant errors, hitting civilian homes even in the absence of confirmed military presence.

To mitigate these risks, the IDF have activated preventive procedures such as:

• Issuing warnings before attacks via social media, SMS, voice messages, and phone calls.
• Drones with loudspeakers and parachuted radio transmitters with evacuation instructions.
• Tactical pauses in attacks to allow civilians to flee.
• Humanitarian corridors biometrically controlled to identify militants hidden among refugees.

The Rafah Example

These measures were also replicated during the operation on Rafah in May 2024, allowing for the early evacuation of large urban areas. According to military sources, this approach has reduced civilian casualties, serving as a reference model for Western General Staffs.