When Heat Alerts Aren’t Enough: From Warning to Opportunity
Case of the Week, Min Wu, PhD · ai-public-health.com
It is 3 PM on a 95°F afternoon. Mr. Smith, in his late seventies, lives alone in a third-floor apartment in a city he has lived in for forty years. The window AC unit died two summers ago and was never replaced. He has cardiovascular disease, mild mobility limitations, and a TV that runs on the local news channel most of the day.
The heat advisory has been on the screen for six hours. He has seen it. He has heard it. He understands it. None of that has changed anything about the next two hours of his afternoon. The advisory told him the air was dangerous. He already knew.
Last week I wrote about the soccer game my friends and I moved from 3 PM to 5 PM, and what that small adjustment meant for the people who couldn’t make it. Mr. Smith is one of those people. The advisory he’s watching wasn’t designed for someone in his position — not because it’s wrong, but because it stops one step short of what he actually needs.
A note before going further: Mr. Smith is a composite. He stands in for older adults I’ve worked with, taught about, and read about — people whose vulnerability the heat advisory infrastructure tends to miss. The specifics of his afternoon are illustrative, not biographical. The gap his story points to is real.
The Difference Between a Warning and an Opportunity
A heat advisory is a piece of risk communication. It tells you the threat. It assumes you can convert that information into action on your own — that you have the resources, the options, and the flexibility to respond. For a substantial fraction of the population, that assumption holds. For Mr. Smith, and for many like him, it does not.
Public health has built a serious infrastructure around risk communication. Thresholds, advisories, color-coded alerts, recommended behaviors. The infrastructure works well when the audience can act on what it’s told. It works poorly — sometimes catastrophically — when the audience is exactly the population most likely to be missed by it.
What Mr. Smith actually needs is different from what the advisory provides. He needs to know that the cooling center six blocks east has capacity right now. That a city vehicle can pick him up in twenty minutes. That his neighbor Mrs. Chen has been notified and will check on him at 4 — and that her electric car, parked in the courtyard downstairs, is available in the meantime; its battery can run the air conditioning for the better part of a day without meaningfully draining it, which makes it a cooled space he can reach faster than either the center or the vehicle. None of that is information about danger. All of it is a path through the next two hours of his day.
That is the difference between a warning and an opportunity. A warning tells you the world. An opportunity tells you what is feasible for you, right now, given who you actually are.
The question isn’t whether people know the heat is dangerous. The question is whether the system can offer them something useful given who they are and what they have access to.
What PAPO Tries to Do
I’ve been working on a framework called PAPO — Policy-Aware Personalized Opportunity — that tries to make this shift structurally. It has four components, and the easiest way to explain them is through Mr. Smith’s afternoon.
The Policy Input Layer encodes what the city’s heat emergency system already permits. Cooling centers are open. Transport is available for residents who meet certain thresholds. Mr. Smith, with his age and chronic condition, meets them. The framework doesn’t replace policy; it reads it, so the rest of the system knows what is institutionally allowed.
The Personalized Opportunity Engine combines who Mr. Smith is with what is currently available. His age, his cardiovascular condition, his apartment without working AC, his location, his preferred mode of contact — all matched against real-time data on cooling center capacity, transport availability, volunteer coverage, and increasingly, shared resources like a neighbor’s parked electric vehicle that can serve as an immediate cooled space while a longer-term option arrives. The engine produces a small set of options that are actually feasible for Mr. Smith, not for the average resident of his ZIP code.
The Opportunity Delivery Interface reaches him through a channel he uses. Mr. Smith doesn’t read text messages well. He answers the phone. The system calls him, in plain language, with the specific opportunity that fits his afternoon. If he doesn’t pick up, the system doesn’t give up — it escalates to a human worker or to Mrs. Chen next door, depending on what the policy permits.
The Feedback Loop records what happened. Did Mr. Smith accept the offer? Did the vehicle arrive on time? Did the cooling center have capacity when he arrived? Was the next person on the list still served? The data feeds back into the system — not to optimize Mr. Smith’s individual experience, which has already passed, but to refine how the next afternoon’s opportunities are matched, delivered, and supported.
These four components are not technically novel on their own. What is meant to be novel is the organizing logic: that the unit of intervention is an opportunity, not a warning, and that the system’s job is to make policy meet personal context rather than to broadcast averages.
What This Doesn’t Solve
I want to be careful about what PAPO claims and what it doesn’t.
PAPO does not predict who will die in a heatwave. It does not replace policy or human judgment. It does not work in cities without basic data infrastructure — without digital records of cooling center capacity, without GIS layers for transport routing, without some baseline of community outreach that a system can plug into. In cities that lack this infrastructure, the framework is a planning conversation, not a deployable system. And when an opportunity depends on a neighbor’s goodwill and property — a parked car, an open door — the framework can surface it, but it cannot guarantee it; consent, access, and liability stay with the people involved, not with the system. That dependency cuts both ways, though. An opportunity like Mrs. Chen’s car doesn’t wait on a Policy Input Layer, a real-time data feed, or a city procurement cycle — it can start working on a single block this week, in a city years away from building any of the rest of this. Some of what people need during a heatwave shouldn’t have to wait for the system that’s supposed to deliver it.
PAPO also does not fix the underlying inequities that put Mr. Smith in a too-hot apartment in the first place. His broken AC, his fixed income, his isolation — none of those are problems an opportunity-delivery framework can solve. A better heat response is not a substitute for housing policy, energy policy, or social policy. It is, at best, a way of making sure that the policy we already have actually reaches the people it was written for.
What the framework does claim is smaller and, I think, more honest. When a city has heat emergency policies on the books and resources allocated to support them, there is often a gap between what is institutionally available and what individual vulnerable residents can actually access. PAPO is an attempt to close that gap, in a structured and accountable way, without overpromising.
The framework is published, documented, and the simulation work is open. It has not yet been deployed in a real city during a real heatwave. The next round of work is about finding where it breaks before someone has to live through that breakage in the field.
The Advisory and the Afternoon
The heat advisory was on Mr. Smith’s TV the whole afternoon. It was correct, current, and aimed at the right audience. None of that helped him.
Somewhere in the same city, a 35-year-old in a well-insulated apartment with central air saw the same advisory and adjusted his afternoon accordingly. The information was identical. The day was not.
Most of what public health does in a heatwave will continue to be risk communication, and that’s appropriate — the people who can act on warnings should get warnings. But for the population that risk communication has historically failed to reach, the next step is asking what an opportunity would look like, and whether we can build systems that deliver it. That’s what PAPO is trying to be. The work is unfinished. The afternoon is not waiting.
As a member of the Springer Nature Author Affiliate Program, I may earn a small commission from purchases made through this affiliated link to my book. Support the author by checking out my textbook, Artificial Intelligence in Public Health: https://tidd.ly/4mH9389
