A study published in Climate Risk Management reveals that artificial intelligence and machine learning research on extreme heat is overwhelmingly dedicated to forecasting risk rather than helping officials respond. The analysis highlights a significant imbalance in how the technology is currently deployed for climate adaptation.
The review of existing literature found that 98.4% of the papers focused on prediction, while 59.2% addressed prevention. In contrast, only 10.4% examined response strategies, 3.2% looked at recovery, and a mere 1.6% covered preparedness. The authors describe this disparity as a "research-to-implementation gap," noting that the field prioritizes identifying where people are in danger over determining how to act on that information.
This imbalance exists against a backdrop of rising mortality from high temperatures. According to a 2025 Yale study, U.S. deaths associated with heat rose 53% from an annual average of 2,670 in the period from 2000 to 2009 to more than 4,000 in the period from 2010 to 2020.
Karishma V. Patel, a RAND policy researcher and co-author of the study, stated that AI adds value where prediction is central to decision-making, particularly in providing scale or granularity. However, she emphasized that building the technology is insufficient if it does not solve a specific problem or remove a bottleneck.
Jennah Gosciak, a PhD student in Cornell's Information Science Department who previously served as a Public Interest Technology Fellow with New York City Emergency Management, noted that creating "robust data systems" capable of ingesting and processing diverse data sources is most of the battle. She distinguished between "slow" and "fast" data, suggesting that simpler heat-vulnerability indexes may be better suited for longer-term planning because they allow officials to explicitly choose which vulnerable populations to prioritize. In contrast, predictive algorithms may be more useful for time-sensitive decisions tied to specific trigger events.
Google has expanded its AI-powered Heat Resilience data from an initial 14-city pilot to more than 50 cities worldwide, an announcement made in June. The system uses AI and high-resolution imagery to map roof reflectivity down to individual buildings, providing urban planners with building-level data. Google stated that data from these pilots helped inform decisions in several cities, including cool-roof ordinances and adaptation plans.
Patel suggested that the next frontier involves connecting predictions to interventions rather than incrementally improving forecasts. She proposed that AI could assist administrators in making immediate choices by synthesizing evolving signals and offering highly localized forecasts at a granular scale. This approach could aid in understanding compound threats involving heat, air pollution, wildfire smoke, and drought.