Data Scientist Confronts Algorithmic Biases and Dispels the Myth of Computational Fairness
True or false? Humans are fallible and biased while computers are logical and unbiased.
For years, data scientists believed in this claim, and algorithms were written with confidence that the results they returned were unbiased. But recent work in data science and artificial intelligence has proven that this is not the case.
“Algorithms reinforce human biases. They amplify them. So algorithmic biases can be much worse than human biases and can have greater impacts,” said Cecilia Aragon (pictured), professor of human-centered design and engineering at the University of Washington.
Aragon spoke with Lisa Martin, host of theCUBE, SiliconANGLE Media’s live streaming studio, ahead of the Women in Data Science (WiDS) 2022 Global Conference. They discussed the importance of recognizing the human side of data science and how Aragon overcame its first “impostor”. syndrome” to become a data scientist, university professor, aerobatic pilot and author.
Five points where humans influence data
There are five key points in the data journey where human data scientists unwittingly influence the data they process. This is discovery, when the dataset is first encountered and used; capture, when the dataset is searched and selected; curation, where the data scientist selects the datasets to be used; categorization; and labeling.
While biases can unintentionally frame the judgments the data scientist makes throughout the process, the last step of labeling is especially critical because judgments here are often subjective, according to Aragon. She gives the example of race labeling based on skin color.
“They may try to apply an algorithm to it, but… we all have very different skin colors, [and] putting ourselves into categories of race diminishes us and makes us less than we really are,” she said.
The secret super power of math
As a Latina who grew up in a predominantly white Midwestern town, Aragon had her dreams of being an astronaut crushed at an early age.
“My teacher said, ‘Oh, you can’t do that. You’re a girl. Choose something else,” she said. She chose math and discovered her super power.
“One of the great things about math is that it’s like a magic trick for young people, especially if you’re a girl or from an underrepresented group,” she said. “Because if you get the correct answers on a math test, no one can prove you wrong.”
But her “impostor syndrome” always kept her from believing that she could obtain a doctorate. in mathematics and computer science. It took becoming an aerobatic pilot to teach her that she could achieve anything she wanted.
“If you’re raised in a way where everyone looks down on you, one of the best things you can do is take on a scary challenge,” she said. “I was afraid of everything. But learning to fly and especially learning to do loops and rolls gave me the confidence to do everything else. Because I thought if I was pointing the plane down at 250 miles an hour and waiting, why would I be afraid to get a PhD. in computer science ?
Ethical and human-centered approaches must be integrated into data science
According to Aragon, understanding the human element of data science is important to mitigate the potentially harmful effects that biased algorithms can have.
“I want to teach the next generation of data scientists and artificial intelligence experts how to make sure their work really achieves what they intended, which is to make the world a better place,” he said. she declared.
Just as built-in security measures don’t provide as strong a protection as built-in measures, algorithm designers must incorporate human-centric and ethical approaches early in the development process.
“It’s not something you slam at the end or say, ‘Well, I’ll wait for the ethicists to step in on this,'” Aragon said. time to think about the impact their algorithm can have.”
Here is the full interview with Cecilia Aragon. WiDS 2022 takes place on March 7. Register here to attend virtually and watch theCUBE’s exclusive coverage of the event.