Divya Gopinath grew up in a household that might seem familiar to many first-generation South Asians – one where math and science were highly valued.
Her father, an engineer, and mother, a neuroscientist, inspired her young mind as she grew up in Chappaqua, New York, in an insular community with few other girls who looked like her.
Gopinath went from being a high school valedictorian to MIT graduate, her resolve and perfectionism seeing to it that she became a machine learning expert addressing the nuance of machine-based bias, and who is now working at Truera, a startup in the Bay Area.
“There were a handful of Indians in the school district I attended, and I was definitely the nerdy math kid,” she says. “It felt restricting at times, like I was wearing the clothes of a stereotype that I was handed. But I also had female figures around me who transformed my worldview.”
She especially credits her mother and her Carnatic singing teacher for acting as role models of working women, and her younger sister, who is her best friend and confidante.
“They never told me there was a limit to how much I could achieve and dream of achieving,” she says.
She was accepted to many top universities, but ultimately chose MIT for its focus on STEM subjects, and quickly gravitated to computer science, specifically machine learning and artificial intelligence. Her research project in her junior year combined computing and neuroscience to build machine learning models to replicate human hearing. She went on to earn her master’s degree there.
At Truera, a small startup in the Bay Area, she grapples with the challenge of safely deploying machine learning. Her master’s thesis, for example, focused on how to leverage machine learning to make clinical note-writing easier for doctors.
Since algorithms can make the difference in high-stakes decision-making, using machine learning equitably is crucial.
“If a human determines to whom to make a loan, then there is an opportunity for them to be biased,” Gopinath says. “If an algorithm were to automate this process, there’s a chance to create a more equitable society that is less bias-prone. But to do that, we have to make sure the algorithm is itself unbiased.”
So how can an algorithm – a set of dry, mathematical operations – become biased?
“Machine learning models are created by mimicking historical data,” Gopinath explains. “In the U.S., we know that lenders have been biased against applicants of color when granting loans. So if an algorithm is trained on these outcomes, this unfairness can leak into the model. It’s all about the potential to encode unfair relationships between a model’s inputs and outputs. Let’s say a model is given access to demographic information, then there’s a potential for it to latch onto biased assumptions for a particular race.”
Even if not explicitly informed, the machine can observe and incorporate the bias in the data.
“Even if the model is not given access to protected data like race, it still has the potential to be unfair,” Gopinath says. “Geographic data – like a zip code – can act as a proxy for protected information because it is inherently correlated with race. An algorithm can still be racist despite having no access to a person’s race [information]. This is why it’s difficult but so important to rigorously and mathematically test a model for bias. Although machines are more efficient than humans, we need to incorporate or instill the human factor into the dynamics of decision making. That’s what I work on.”
Gopinath credits her interest in this field to her experiences as a female engineer. She notes that in more mathematical fields like computer science and, specifically, machine learning, there are far fewer women than men. The corresponding dynamics has informed her experiences, and affected her career path.
While working on a group project at MIT, one male classmate told Gopinath that he would handle the math for the assignment and that she could work on the visual design.
“I value humility, but I never want anyone to think I’m unfit to do technical work or to put me in a box I don’t want to be in,” Gopinath says. “So I told him that his proposed arrangement wasn’t going to work. I didn’t let him define who I want to be.”
She is proud of that moment because it set her up for later experiences at MIT.
“In many of the more mathematical graduate-level courses I took, I was one of two or three women in a lecture hall,” she says. “That’s disheartening, but you can’t let it deter you from learning about what you love… Women can have stellar resumes and impeccable pedigrees, and their qualifications will still be debated by society because they’re women, or women of color. Don’t buy into that mentality, for yourself or for your female friends. You did the work and you are qualified. Your gender doesn’t matter.”
But she points out that bias is an integral part of human psychology – and finds its way into their creations.
“We’ll never be in a place where we can completely trust humans or algorithms,” Gopinath says. “Humans have a tendency to place unnecessary labels on people. … I’ve seen this both as the sole South Asian kid or female engineer in a room. If we’re careful, we can certainly teach machines to be better, but like humans, they’ll never be perfect. That imperfection intrigues me, for the goal post is always moving.”
Gopinath says she’s lucky to have grown up without being held back, and that one of her biggest challenges is living up to expectations about her.
“I grew up with people telling me that I would change the world today, but I’m realizing that I’m unlikely to achieve this via a single discovery or big breakthrough,” she says. “It’s about very small incremental progress towards something bigger.”
She also concedes it is important to shrug off those expectations.
“To some extent, I have followed a path that was laid out for me,” she says. “I studied the right things and checked the right boxes. And yes, I’m passionate about computer science, but that’s not all I am. I want people to see me as multidimensional.”
Gopinath is a singer herself, and constantly listens to music to center herself.
“It’s something I can appreciate and remind myself that I am a person beyond my laptop,” she says.
Back in her last semester of high school, she had captivated the audience with her voice in a play.
“That musical was a defining moment for me,” Gopinath says. “I had been studying classical Indian vocals for over a decade but I didn’t tell my classmates in school because I was afraid of the cultural barriers. But when I finally got an opportunity to sing for them, the community saw me as more than a stereotype. It was empowering.”
She adds: “Once I came to terms with the fact that I was entitled to be both a musician and an engineer rather than choose a singular role, I didn’t look back. I directed one of MIT’s a cappella groups for a few years in college and was in a vocal jazz group as well. I still compose music in my free time.” Asked about her proudest moments, she cites small victories rather than technical awards.
“In the a cappella group I was in, we worked really hard for a few competitions, and it felt so validating to be able to perform at that level. That was a moment of triumph,” she says. “There will always be research accomplishments or career milestones, but this was a reminder that I can do things outside of a classroom, too.”