By Amanda Gillooly, GASP communications manager

I’ve got a confession to make: The hard sciences aren’t really in my wheelhouse. Chemistry. Physics. Basically, anything that deals with taking measurements or understanding spacial relations. Not my thing, man. I’m but a humble writer.

So, when my boss here at GASP asked if I’d like to attend a training event that would teach me how to properly label smokestack emissions from the Clairton Coke Works, my immediate thought was, “I’m probably not the best candidate.”

Then she told me the end goal: That these volunteers would actually be helping to train artificial intelligence to detect industrial smoke pollution automatically.

In my head, there was a record screech. “I’m totally gonna bring down the curve,” I thought.

But I was willing to give it the old college try.

So that night, I joined a few other volunteers and our friends from Carnegie Mellon University’s CREATE Lab, which developed the tool and would be conducting the training.

The developer told us where to log in, and made some introductory remarks. He reiterated that we would be seeing video clips that depicted emissions from various industrial sources. He explained that the big issue is discerning whether the plumes are steam or smoke.

He added that it would take about 300 hours to amass enough correctly labeled smoke videos to train AI to automatically identify it.

It took all of five minutes for me to blurt out, “What if I just don’t get it and label steam as smoke. Won’t I actually be doing a disservice to this whole endeavor?”

That’s when the friendly (and really, really smart) folks at the CREATE Lab told me about safeguards set up for this very reason. They assured me that people who incorrectly label enough videos would be asked to go through prompts providing additional instruction.

“Don’t worry,” Ana said. “I have complete confidence that you will be able to do this.”

That made exactly one of us. But heartened with the knowledge that I wasn’t gonna mess the whole thing up, I began labeling videos.

And immediately screwed up enough that the system politely suggested that I take the tutorial again.

“Naturally,” I thought as I scrolled through the additional instructions, only to incorrectly label a bunch more videos.

This is the point that I would normally want to flip a proverbial table.

But let me tell you: The young man who developed the program, Yen-Chia Hsu, truly created a stellar tutorial—one that congratulates you when you get something right, and explains in a friendly, helpful way what you got wrong.

After trying and failing and trying again, I was successfully able to label smoke for the next hour or so.

My point is: If I can do it, you can do it. And you can do it from your phone, or tablet, or laptop. At your own convenience. And in your own home.

“Even if people did it for 10 minutes a day, it would be a huge help,” Yen-Chia said.

By this point, I hope you’re telling yourself, “Sounds like a pretty easy way to make a difference. I’m into it – so sign me up.”

If so, https://smoke.createlab.org/ – go check it out and start smoke reading today.