Researchers are turning to AI to give workers the best chance of finding dangerous lead in Flint.
Artificial Intelligence is Helping Get the Lead Out of Flint
Published: November 19, 2019
Gina Luster: We could be getting water from heaven but our infrastructure is so destroyed, by the time it makes it to me and my family, it’s no good.
Onscreen: In 2014, Flint, Michigan changed water sources but didn’t properly treat the new water to prevent corroding water pipes resulting in eroded lead pipes and contaminated water across the city
Luster: I will never trust what anyone says as far as the condition and quality of the water. In the back of my head, I'll always be wondering.
Onscreen: Officials denied that the water was unsafe for more than a year until activism by Flint residents brought national attention and pressure.
In 2016 the city began to replace the lead pipes. But there was a huge problem. They didn’t know where the thousands of lead pipes were.
Jacob Abernethy: The city had essentially no records of what these lines were. The vast majority of the lead pipes were in the public portion of the line.
The problem is that the line is buried entirely underground. There is no way to find out what the material is unless you dig six feet underground and that’s very expensive
Onscreen: To give workers the best chance of finding the lead, U. of Michigan researchers turned to a form of artificial intelligence (AI), that searches large collections of data looking for patterns. It’s called “machine learning”
Abernethy: It first would check, is the age of the property above this value? Then, is the home located in this neighborhood? Then, has this home had a water test?
Onscreen: Using census data and city records, the algorithm makes conclusions based on relationships it finds in the data.
Abernethy: It takes all these little pieces of information and combines it into a single number which is a probability of a home having a dangerous pipe.
Onscreen: The more information the ai model got about pipes in Flint, the more accurate the predictions became.
Abernethy: When this project started, we had a total of 36 homes where we actually knew the service line material.
Now we have over 20,000 homes. Once you go from 36 to 20,000, the accuracy of a machine learning algorithm increases dramatically.
Onscreen: The city of Flint started using the AI model to prioritize digging at homes that most likely had lead pipes. In 2016-2017, the algorithm successfully predicted which homes had lead about 81% of the time.
Abernethy: That was a decent hit rate, it wasn't perfect.
Onscreen: But many residents objected to prioritizing some homes over others.
Maurice Davis: When it's something as devastating as the lead in this water, we don’t need predictions, we need 100% accuracy to right this wrong.
We need to actually dig holes, excavate the holes, look down in that hole and see copper-to-copper or lead-to-copper. You come and replace my neighbor’s but you don’t replace mine according to a prediction? That don’t help nobody’s confidence.
Onscreen: In 2018, the City hired a California-based contractor to accelerate replacement of lead pipes.
They stopped using AI to target specific homes and took a more blanket approach, inspecting neighborhoods across the city.
Abernethy: They did a less targeted, less focused approach
Onscreen: But only found lead in about 15 percent of the homes inspected
Abernethy: When it became clear that the hit rate of this new contracting firm had gone down so dramatically, people were concerned about that and were trying to figure out why finite government resources were being used to try to replace pipes in homes that actually didn't require replacement.
Onscreen: A coalition of activists sought a court order demanding that the city first inspect homes expected to have lead.
A settlement required the city to use the AI model. Now the hit rate has climbed back up, to over 70%. Still, many residents want physical confirmation that their pipes are safe.
Luster: Once all of these agencies that are supposed to be put in place to protect you deceive you and lie to you, how do you ever gain that trust back?
The damage is permanently done to our bodies, our infrastructure. You can never forget that. Our pipes haven’t been changed yet. I don’t really know what our service line pipes are. I want a person down there with a shovel, literally down there touching that pipe.
Eric Schwartz: There’s really no substitute for seeing someone dig in your front lawn, removing that lead pipe, and replacing it with a fresh copper pipe.
The goal is really just take advantage of all the information we have and remove the lead from the homes that actually have lead as quickly as possible so that no residents in Flint have to continue living with lead more than five years after the water has been switched.
Onscreen: More than 8,000 lead pipes have been replaced so far.
No one knows how many remain. Researchers estimate there could be about 1,921 lead pipes remaining
Crowd: Five years later, Flint is not fixed
E. Yvonne Coleman-Lewis: There are things that will never change in my life or my family’s life as a result of this.
Until the water system is completely replaced, there will be little, if any, true trust in the water that comes from the tap.
Onscreen: At least 12 residents died from illnesses related to the water crisis in Flint
Crowd: Rest in peace. All of you fallen soldiers. That’s what you are.
Onscreen: It’s estimated that in 2014-2015 alone, 140,000 people were exposed to dangerous levels of lead in Flint’s water including around 9,000 children.
Produced and Edited by: Emily Zendt
Produced and Directed by: Carol Yancho
Camera: Vicente Franco, Jason Vlahos, Hannah Byrd
Sound: Mark Hayge, Drew Hill
Animation: Mitch Butler, Diego Arenas
© WGBH Educational Foundation 2019