New Jersey Transit incorrectly used federal weather prediction software to model Superstorm Sandy's storm surge, resulting in severe underestimation of flood risk that led to hundreds of millions of dollars in equipment damage when trains were not moved to safety.
In October 2012, New Jersey Transit used National Weather Service storm prediction software to help decide whether to move trains from its Meadows Maintenance Complex in Kearny, New Jersey before Superstorm Sandy struck. According to documents obtained by Reuters, agency officials incorrectly configured the software by modeling a storm traveling northeast at 10 miles per hour, when Sandy was actually moving west-northwest at 20 mph toward the New Jersey coast. This error caused the software to predict storm surges of about 3 feet at The Battery in Lower Manhattan, when federal forecasters were warning of surges up to 8 feet. The actual storm surge reached 9.23 feet. As a result of this miscalculation, New Jersey Transit left hundreds of rail cars and locomotives worth hundreds of millions of dollars in low-lying rail yards. The storm surge ultimately flooded the yards, causing $100-120 million in damage to one-third of the agency's locomotives and a quarter of its passenger cars - over a quarter of the entire fleet. This led to several weeks of service disruptions with long delays and crowded trains for New Jersey residents commuting to New York City. National Weather Service meteorologists confirmed the software settings were wrong and said they could have provided assistance if contacted.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
Human
Due to a decision or action made by humans
Unintentional
Due to an unexpected outcome from pursuing a goal
Pre-deployment
Occurring before the AI is deployed