Faculty members in Duke’s Computer Science and Electrical & Computer Engineering departments work to promote equal opportunity in their respective fields.
Fairness problems in computer science can arise in the development of software at a fundamental level. Just as an author’s beliefs and implicit biases can be revealed in the literature, how a programmer creates a predictive model can be shaped by his own beliefs and biases. The lack of diversity in this area means that algorithms can inadvertently perpetuate stereotypes.
Dubbed algorithmic bias, this phenomenon can have serious consequences as society relies on predictive models in a number of critical areas. For example, a 2019 study found that an algorithm used to estimate future medical costs for over 200 million US hospitalized patients favored white patients over black patients.
According to Jian Pei, a professor of electrical and computer engineering, it is crucial that high-stakes models do not further disadvantage marginalized subgroups.
The deconstruction of these prejudices was Pei’s main task. In combating algorithmic bias, Pei takes a three-pronged approach: raising awareness, effectively assessing bias, and implementing equitable technology.
“We should use case studies and examples to detail the potential for a lack of fairness, diversity, and equity in algorithm design. If people are aware of these issues, the next step is how we can measure them,” Pei said. “People don’t know how to use these concepts or how to approach these concepts. It is important to develop quantitative measures such as B. how unfair or how close it is to being fair [a model is]and what the possible compromise is.”
Pei explained that fairness should be considered at several stages of the data science lifecycle, including how data is collected, sampled for analysis, and evaluated for post-processing. It is not enough to adjust outcomes that appear to benefit certain subgroups over others; Instead, models should be designed to perform unbiased calculations.
“We need to be proactive and build mechanisms to avoid future problems. Not only does this fix an existing problem, we need to improve the entire process going forward,” said Pei.
Pei is not alone in his efforts. Cynthia Rudin, Earl D. McLean, Jr. Professor of Computer Science, Electrical and Computer Engineering, Statistics, and Biostatistics and Bioinformatics, has emphasized interpretability and transparency in her work. Rudin has received recognition for her research examining bias in black box models used in healthcare and criminal justice—algorithms whose internal workings are hidden.
“In criminal justice, we’re trying to show that black box models can be replaced with interpretable models without loss of accuracy, making the models less susceptible to the invisible biases and data entry errors that black boxes have,” Rudin wrote in an email.
As a professor, Rudin has approached justice in the classroom from multiple angles. For example, she has made her courses more accessible by putting materials online, both for students who are unable to attend classes and for the general public. Rudin also updated her curriculum over the past year to include interpretability as a key concept.
“Once we start working on something important where error plays a role, like health care or credit decisions, it’s really important if we understand what the predictive models are doing,” Rudin wrote. “I’ve decided that if you want to become an expert in machine learning, you really need to know something about interpretability.
Rudin said her students are “the first students to actually be taught this material in a classroom anywhere in the world.”
Get The Chronicle straight to your inbox
Sign up for our weekly newsletter. Cancellable at any time.
| staff reporter
Gautam Sirdeshmukh is a Trinity senior and a staff reporter for the news department. He was previously the Health and Science News Editor for Volume 117 of The Chronicle.
