Computer Scientist Nisheeth Vishnoi Warns that Algorithms Reflect Human Biases
Convening Yale
Wednesday, Sep 23 2020 at 11:45 am - 12:45 pm EDT
Please join us for the first Dean's Office Speaker Series event of the semester with Professor Nisheeth Vishnoi on the “Challenges of Making Unbiased Decisions in a World of Data and Algorithms.”
A combination of biased data and decontextualized models has led to a host of “intelligent” algorithms that are capable of reflecting and exacerbating human biases. As the use of such algorithms in societal contexts proliferates, the question arises: How can we redesign them to mitigate bias?
The conversation will be moderated by Shyam Sunder, the James L. Frank Professor of Accounting, Economics, and Finance.
Convening Yale presents talks by faculty and leaders from throughout Yale University, who share their research and expertise and help students broaden their understanding of an increasingly complex world. The Convening Yale series is made possible through the generous support of the Robert J. Silver ’50 Fund for Innovation in Management Education.
Algorithms are not the objective purveyors of truth that many people imagine, and their curations and conclusions should be greeted with a healthy dose of skepticism.
That was the message that Nisheeth Vishnoi, professor of computer science, shared with a Yale School of Management audience in a virtual event on September 23.
“Algorithm design is a human-driven process,” Vishnoi said. “Despite being governed by precise mathematical rules, Algorithms can be very brittle, and human biases can creep in, and they’re not really as objective as people think.”
Vishnoi’s talk, “Challenges of Making Unbiased Decisions in a World of Data and Algorithms,” was the first event of the fall in the Convening Yale speaker series, which brings experts from across the university to share their research with Yale SOM students. Shyam Sunder, the James L. Frank Professor of Accounting, Economics, and Finance, moderated the discussion.
Vishnoi described the ways that biased data and decontextualized models build inaccurate distortions into algorithm functioning. With algorithms playing a growing role our lives, redesigning them to mitigate bias is critically important.
“We are surrounded by societal technology today, governed by algorithms, which learn from our behavior and make models of us,” he said. Algorithms help dictate decisions on everything from who gets bank loans to what items show up in social media newsfeeds, Vishnoi said.
A “vicious cycle” ensues when human bias regarding the collection of data infects algorithm design, and algorithms then perpetuate the cycle by aggregating more skewed data and presenting it to people as “reality,” Vishnoi said.
The process is particularly destructive in a democratic society, where the flow of unbiased information is critical in allowing voters to accurately assess issues and political candidates.
“People’s perception about reality is fundamentally becoming more polarized,” Vishnoi said. “When algorithms become distorted, they alter human perception.”