First, many countries have their own laws and policies regarding air quality exposure. Often times these regulations use metrics about time-average exposure (e.g. the annual average 8-hr maximum ozone concentration) and can't trivially be converted between agencies. It causes very negative user experiences when your air quality summary disagrees with what a regulatory agency is putting out to the public due to a measurement technicality - much easier just to localize the data in the same way you'd localize the rest of the UI.
Second, we're stuck with AQI and similar indices because air quality is a multi-faceted problem with different components measured wildly differently. For instance, PM2.5 and PM10 are typically reported in ug/m^3 - but who actually knows what these measurements mean? What's the difference between 15 and 30 ug/m^3? It's very far outside of the experiences that many individuals have, so end users really struggle to interpret these values, even when you translate them into simple "low"/"medium"/"high" categories. Then of course gas-phase pollutants are typically reported in volume concentrations like parts-per-billion or parts-per-million. Aside from the fact that these values are very commonly misinterpreted as much smaller than they actually are, the difference in units adds another bit of complexity into users interpreting what they're seeing.
A well-designed air quality index could factor our all these challenges. I definitely h/t Breezometer for their attempts there, even if it is generally poorly received. But it's a very challenging problem no matter how you slice it.