My team's new AI model totally misjudged a wildfire risk in Colorado last month

We were testing our new prediction model on some real data from the Rocky Mountain area, and it gave a low risk score for a zone that actually had a big fire a week later. I mean, we had fed it years of weather info, but it missed some key stuff about the dry ground. Has anyone else had their AI mess up on real world environmental data like that?

5 comments

5 Comments

angela_robinson3mo ago

We added soil moisture sensors and it finally started catching those dry ground risks.

anna_coleman463mo ago

But what if the weather data was just wrong?

rileygarcia3mo ago

I used to think weather data was enough for those models. But seeing what happened to you and reading angela_robinson's fix, it's clear the ground itself tells you a lot more.

davidkim2mo ago

My garden taught me the same lesson last summer. The forecast said rain but the dirt was bone dry, and my tomatoes suffered for it. You really can't ignore what's happening right under your feet.

jamie8042mo ago

You're totally right about the ground telling the real story. My plants have been saved more than once by checking the soil myself. Data on a screen just doesn't compare to sticking your finger in the dirt.