Why We Still Need Girls to Code
By Dr. Darren Kessner, STEM Program Co-Head and Computer Science Instructor
In her 2018 research, Joy Buolamwini demonstrated that commercial facial recognition software from major tech companies was much more successful at identifying white male faces than brown or female faces. In a more recent study on the faces of U.S. members of Congress, men were tagged with labels such as "official" and "businessperson", while the most common labels associated with female faces included "smile" and "hairstyle".
There is no doubt that the engineers, project managers, and chief executives of the companies involved in these projects were hard working and had no ill intentions. How is it possible that these companies have released software that produces biased output?
The answer is clear: there has been a historic lack of diversity of voices in the development of these software projects, from research to development to testing to public release. If the CEO of one of these companies were a Black woman, it would be highly unlikely for the company to release facial recognition software that did not recognize her face.
Artificial intelligence (AI) powers a huge amount of software, for example facial recognition, entertainment recommendation, targeted ads, fraud detection, risk assessment, language translation, and voice assistants. Software engineers train AI software algorithms with data sets, and lack of diversity in the data can lead to bias in the algorithms. In addition, any patterns in the data caused by historic inequities will be perpetuated if the algorithm is used for prediction or recommendation.
Biased software is a problem for two reasons: 1) people from underrepresented groups may have less access to important services, and 2) policy relying on such software are likely to perpetuate any patterns of historic inequity.
The Computer Science curriculum at Marlborough has grown over the past 6 years under the principles of open access and the promotion of diversity. First of all, there is no bar to entry: our introductory coding classes have no prerequisites, and we welcome all students.
We use open source software and open access educational resources for our computer science classes. All of our classes are project-based and students are given the support and opportunity they need to master the skills required to create software. In addition, we encourage civic engagement through the analysis of publicly available data sets, both in coding classes and for research projects.
Finally, we encourage our students to explore applications of software in a wide variety of areas, such as math and science, humanities and social sciences, and the visual and performing arts. Our coding students create websites, video games, interactive data visualizations, mobile apps, wearable electronics, robots, generative visual art, and video projections for dance performances.
Our future will be determined by who creates software and how that software is used. The goal of our Computer Science curriculum at Marlborough is to promote a future where diverse teams of people ensure that the software we are creating is unbiased, and where diverse teams of people make decisions about how we use that software in our society.