AI summary by TwinMind
How AI tells Israel who to bomb
Vox
9m 13s
Personal Impact - Heba's Story
- Northern Gaza resident evacuated October 11th, 2023
- Home destroyed by February 2024
- Family had trees, flowers planted - all gone
- Speaking from friend's home in Rafah, southern Gaza
- Represents broader pattern of AI-directed destruction
Israeli AI Military Systems Overview
- Iron Dome: defensive system against missile attacks
- Defended against Iran's drone/missile attacks April 2024
- SMASH (Smart Shooter): AI precision assault rifle sight
- Uses advanced image processing algorithms
- Auto-aim functionality like Call of Duty
- Palestinian surveillance in occupied territories
- Facial recognition at hundreds of checkpoints
- Biometric matching against databases
Gospel AI System - Building Targeting
- Produces bombing targets for specific buildings/structures in Gaza
- Works with other AI tools for data collection
- Uses surveillance and historical data on Palestinians/militant locations
- The Alchemist platform: collects/transfers data between departments
- Fire Factory: observes data, categorizes it
- Four target categories:
- Tactical targets: armed cells, weapons warehouses, launchers, headquarters
- Underground targets: tunnels under civilian homes
- Family homes of Hamas/Islamic Jihad operatives
- "Power targets": residential/high-rise buildings with dozens of civilians
- Power targets designed to exert civil pressure on Hamas
- First 5 days of war: 50% of targets were power targets
- Produces targets faster than humans can process
Lavender AI System - Individual Targeting
- More secretive system for targeting specific people
- Generated up to 37,000 Hamas/Islamic Jihad targets using historic data
- 10% error rate according to sources
- Expanded definition of Hamas operative to include:
- Police force
- Doctors
- Civil society workers (anyone interacting with Hamas governance)
- Links targets to specific family homes
- Recommends weapons based on operative ranking
- Lower-ranking militants targeted with "dumb bombs" (unguided, cheaper)
- Creates more collateral damage due to cost considerations
Civilian Casualty Thresholds
- Junior Hamas operatives: permissible to kill 15-20 civilians
- Some targets: permissible civilian casualties up to 300
- Over 34,000 Palestinians killed vs 1,400 in Israel's 2014 Gaza war
Human Oversight Limitations
- AI systems produce predictions, not facts
- Quality dependent on human creators' understanding
- IDF claims human analysts conduct independent examinations
- Actual human supervision for Lavender targets:
- Single check ensuring target is male rather than female
- No other verification protocols reported
International Framework and Accountability
- November 2023: US released international framework for responsible AI use in war
- 50+ countries signed
- Israel has not signed treaty
- Lack of sufficient oversight and accountability
- Gaza described as "unwilling test site" for future AI technologies
Future Implications
- Trend toward more imprecise and biased automation
- Faster warfare does not lead to global security/peace
- Need to interrupt momentum of these technological initiatives
- Civilian casualties expected to worsen with expanded AI targeting
