Stony Brook, NY, December 20, 2025 — Arizona’s blistering summers are notoriously extreme: record-breaking temperatures, workers at risk, crowded emergency rooms. In a study led by Stony Brook University researcher Jieshu Wang, her team asks a pressing question: what if the people most exposed to extreme temperatures are also the ones working against climate change?
Wang, an assistant professor in the Department of Technology and Society in Stony Brook’s College of Engineering and Applied Sciences (CEAS), used large language models (LLMs) to help map how Arizona’s workforce contributes to “heat resilience” — the ability of people, infrastructure, and institutions to withstand and adapt to extreme heat. The study was published in the journal Energy and AI.
“Previous research often frames workers as victims of heat,” Wang said in an interview. “But our workers are a brilliant human resource that can help us build heat resilience. Heat-smart housing, green infrastructure, transportation — everything we rely on in extreme heat is built and maintained by people.”
The study grew out of Wang’s earlier work at Arizona State University’s Decision Theater and Knowledge Exchange for Resilience (KER), which was supporting the governor’s office on extreme heat preparedness. Arizona is one of the hottest places in the United States; in 2024, Phoenix recorded 70 days above 110°F and 113 consecutive days at or above 100°F. These conditions also made it an ideal test bed for understanding how labor, climate, and infrastructure intersect.
Wang and her colleague, Dr. Patricia Solís, KER’s executive director and a research professor in ASU’s School of Geographical Sciences and Urban Planning, turned to the U.S. Department of Labor’s database of occupations and the tasks these jobs involve. For Arizona alone, the database listed 663 occupations and 16,398 distinct work tasks, from installing air conditioners to drafting legislation.
Manually reading and rating each task for its relevance to heat resilience would have been impossible. Instead, the team instructed GPT-4o to score every task on a five-point scale based on how much it helps a community prepare for, withstand, or recover from extreme heat. The model also assigned each task to one of several “heat-solution roles,” including engineering and technical work, health and safety, environmental research, or public service and emergency response.
“Using a large language model with carefully designed prompts turned out to be a pretty efficient way to automate this kind of human reasoning,” Wang said. “We can ask it how much each task contributes to heat resilience, and have it explain why!”
To keep AI in check, Wang iteratively refined the prompts and validated the results manually. She and Solís randomly sampled a subset of tasks, assigned them ratings, compared their scores with the model’s output, and revisited their ratings through further discussion informed by the AI’s explanations. Over time, agreement between humans and GPT-4o rose to 83%, with correlations in line with — or slightly better than — previous benchmark studies.
The results reveal a surprising picture of who is already on the front lines of heat adaptation. Of Arizona’s 663 occupations, 110 were identified as “heat-solution occupations,” meaning a substantial share of their day-to-day tasks help reduce heat risks or improve resilience. Together, those jobs employ about 435,000 workers, i.e., 14.36% of the state’s workforce.
These occupations range from the expected to the unexpected — HVAC technicians, solar installers, insulation workers, and roofers clearly help keep buildings livable and energy-efficient in extreme heat. Environmental scientists, hydrologists, and atmospheric researchers provide long-range data and models that help communities understand how heat is changing over decades. Healthcare workers — paramedics, nurse practitioners, and community health educators — respond to heat-related illness and educate the public about staying safe.
Then there are the “dual-impact occupations”: 31 jobs that are both highly exposed to heat and critical to heat solutions, employing roughly 176,570 workers, or 5.8% of Arizona’s workforce. Electrical power-line installers, landscapers, tree trimmers, and iron and steel workers fall into this category. Many spend their days outdoors in triple-digit temperatures in summertime, yet their work directly affects how well communities weather heat waves.
“When we looked at these dual-impact occupations, we really felt we needed to protect them better,” Wang said. “They’re such a valuable human resource for the region.”
One of the most striking findings was the role of legislators. With GPT-4o’s assistance, the researchers classified more than half of legislators’ listed tasks as contributing to heat solutions, including drafting laws and policies that shape housing, energy use, and public health. Wang points to a 2024 Arizona law guaranteeing mobile-home residents the right to install cooling measures — a change that potentially helped drive the first decline in heat-related deaths in Maricopa County in over a decade. “It really shows how researchers and lawmakers working together can change the conditions people live in,” Wang said.
The study has already moved beyond the page. Its framework and findings helped inform Arizona’s first Extreme Heat Preparedness Plan, which calls for expanding weatherization and energy-efficiency jobs and commissioning a study of future workforce needs in key industries. By revealing which occupations matter most for resilience, AI helped offer state and local governments a plan for targeting training, safety regulations, and infrastructure investment.
At the same time, Wang is clear-eyed about the limitations of relying on commercial AI models. LLMs can “hallucinate” or give inconsistent answers, and there is no simple ground truth when researchers ask them to judge nuanced social questions. She argues that the next step for academia is to develop shared guidelines and benchmarks for using LLMs as research tools so their results can be trusted with greater confidence.
Now based at Stony Brook, Wang is eager to extend this work. She sees potential to apply similar approaches to other climate hazards, and to explore how AI is embedded in a wide range of climate technologies, from solar power to carbon capture. She also hopes to pair future AI analyses with fieldwork, interviewing workers in key occupations to understand their lived experience of extreme weather.
“I’m interested in how AI can help us mitigate and adapt to climate change,” she said. “But at the end of the day, the question is human: how can we better use our workforce — and our society as a system — to combat the challenges we’re facing?”