AI, Air Quality, and Health
We can use AI to better prepare for the health effects of poor air quality
Orange air. The smell of burning. Requests to stay indoors with an air purifier—and mask if you can’t.
Even if you weren’t there, you probably heard about it. This was New York City when Canadian wildfire smoke unexpectedly enveloped the northeastern U.S. a few months ago.
As historic wildfires spring up more frequently across the globe, many people who live outside areas typically impacted by these disasters have been learning how to monitor and respond to AQI alerts for the first time.
During New York’s few days of historically poor air quality, social media filled with tips on making DIY air purifiers and recommendations for AQI monitoring websites. Reminders rang out about the hazardous health effects of breathing in such polluted air—comparable to being a serious smoker.
And in the days that followed, criticism of policymakers rang out: We could have predicted this, so why didn’t we warn people?
While wildfires like those in Canada this summer—and the recent historic Maui fire—may be unpredictable, here’s what we know:
We can foresee the flow of degraded air from these and other pollution sources.
Air pollution is incredibly dangerous—current levels have already reduced average life expectancy worldwide by almost 2 years.
So, what can we do? Our thought: Get AI on the job.
As forecasting analytics improve with the help of artificial intelligence, we’re hopeful that this data can be strategically applied to mitigate some of the disastrous health effects of poor air quality.
Step 1: AI-enabled air quality monitoring
One of the coolest parts of air quality monitoring is that the data is abundant.
On any day, you can look up the PM2.5 concentration of your area and, odds are, you can find it. Along with the readings from most major cities around the world.
Go ahead, get Googling. We’ll wait.
Despite how cool it is that you can have air quality readings from around the world at your fingertips, the truth is not quite that simple.
The problem of air quality monitoring hinges on there being both a lot of data—and too little actionable data.
Let’s take the example of one common source of air pollution: Ground-level ozone. You’ll know this bad guy as the main ingredient in smog
The German Tropospheric Ozone Assessment Report (TOAR) database contains data from over 14,000 air pollution measurement sites around the world. And yet—despite that huge number—the data is still incomplete. TOAR data is supplemented by satellite measurements, yet these don’t occur frequently—or precisely—enough for a complete picture.
Plus, a lot of other ground-level AQI data-gathering is citizen science-led. And while a cool initiative, this makes it inherently uncomprehensive.
AI can help fill in the gaps in air quality data by making more “messy” data usable. The Catch-22? We need better data to create well-trained models.
The next level: Predicting air quality hazards with AI
This is where AI can really shine.
But first, a disclaimer: AI-enabled air quality forecasts aren’t entirely new.
This approach to forecasting local air quality first began in the 1990s—when AI was much less powerful than it is now. So, as you can imagine, accuracy wasn’t fantastic.
But now, with AI’s high capacity for tasks like image recognition, it’s time to get excited again.
Many studies have demonstrated that AI-enabled air quality forecasting can actually produce accurate results. And it allows for hourly forecasting with much less computing time. So you can refresh that forecast to your heart’s content.
So, with more complete data, we’re looking at a future where highly accurate and hyper-local AQI forecasts may be a part of any person’s day. Like checking the weather about whether to pack an umbrella for a long weekend.
But how do we actually prepare people?
Once we have this information with the help of AI, what do we do with it?
More detailed air pollution maps can result in much more accurate and actionable targeted mitigation measures. And roll it into preventive care recommendations and routines we already have—like for flu or allergy season.
One of the most important ways we can do this is by helping more vulnerable populations respond to poor air quality. A (so far) theoretical application is the direct measurement and prediction of inhaled pollutant doses that can help vulnerable people better plan their outdoor activities.
And yes, if you’re thinking it, we are too.
We can also use this more targeted, tangible information on air pollution to pressure polluters to reduce emissions.
If you’ve been paying attention, you probably noticed that 2023 just keeps breaking heat records. Again and again.
The writing’s on the wall (and the wall is on fire). Global warming is going to keep creating more of these air quality hazards throughout more of the world.
Plus, many of the most common sources of hazardous air quality are directly manmade. Let’s not forget that, in many cities around the world, AQI readings like those seen in parts of the U.S. impacted by wildfire smoke are an average occurrence.
That is a public health tragedy.
Along with doing what we can to stop and reverse the effects of climate change and fossil fuels, we need to use the tools at our disposal now. AI is part of that arsenal.
And let’s not forget: Yes, you are reading this feature in a newsletter dedicated to news about AI in healthcare. This is where the healthcare part comes in.
When addressing the population effects of climate change, we need to reorient our understanding of what “counts” as healthcare.
Given poor air quality’s severe impacts on human health, we think attempts to mitigate it are well within the realm of healthcare. Let’s step up to the plate.
Let’s think it through: How can the worlds of healthcare and air quality monitoring better work together to protect people’s health? Where do you think AI fits in best?