Warren Center Affiliate Profile- Bo Waggoner
Building a Better (Behaved) Algorithm
What do website ads, medical treatment options, weather predictions, and qualifying for a loan all have in common? All of these things, and more, are shaped by machine learning algorithms, which use large amounts of data to make predictions or decisions.
The widespread collection of electronic data in every area of our lives has led to growth in the application of machine learning. While this growth has allowed for exciting new possibilities in technology, there are critical concerns about the fairness of the decisions that algorithms are making and the disregard of privacy during data collection.
Given the impact that algorithms have on our day-to-day lives, researchers have recognized that we need to reevaluate our understanding of how these algorithms behave and how they get their data.
That’s where Bo Waggoner comes in. Bo’s work as a Postdoctoral Fellow at the Warren Center looks at the privacy and fairness of algorithms, specifically focusing on their impact on people and society.
The Privacy of Data Collection
Users can potentially be providing more information about themselves than they realize as they interact with the world around them, and possible harm or long-term consequences are not always clear. Data can be pulled from web browsers, GPS locations, social media, and even medical records. For this reason, researchers are exploring new and improved ways of collecting information with individual privacy in mind.
Bo discusses his work on privacy: “One of our current projects addresses protecting privacy of personal data that is collected over time from our devices, such as web browsers. This data is incredibly useful, but extremely personal, so we want to ensure that even long-term data collection won’t leak sensitive information.”
Certain companies are have already taken steps to protect users’ privacy, implementing a differential privacy technique. This kind of technique adds noise to the data it collects from individual users, which allows an algorithm to analyze the usage patterns of a large number of people without compromising individual privacy. Apple uses the differential privacy technique in its iPhones, as does Google in the Chrome web browser; it will even be used in the 2020 U.S. Census.
Bo’s work also examines the fairness of the decisions made by algorithms, particularly taking into account ethical considerations and possible biases from imperfect design or imperfect data. While the average user might not be aware of the application of machine learning algorithms, the decisions made by these systems are significant, and have the potential to discriminate against gender, race, or any other subset of people.
For example, loan applications for applicants of different races may be treated differently. As an algorithm takes in large amounts of data, it analyzes which factors makes a person more or less likely to pay off a loan. It then creates a model that will weigh these factors of a particular applicant and then produce a decision. In this way, an algorithm might deny an application based on the applicant’s home address, neighborhood, or even their name.
In order to prevent these kinds of unfair outcomes, researchers are not only exploring how current algorithms learn and make decisions, but also how we can change things to make future algorithms better. In this vein, Bo’s research takes a theoretical approach from a variety of angles to design and analyze systems that acquire and aggregate information.
Moving Towards a Better Future
While conversations about data and algorithms can often feel heady to those outside the field, it’s important to understand the impact they have on our lives. Algorithms live in a social context; data is about people and comes from people. Technical research towards algorithms designed with privacy-preserving statistics and fair decision-making in mind translates into practical consequences.
Bo envisions his research as helping to build toward a future where data science can still be impactful, but with algorithms with strong personal protections and ethics built in. Just as current ideas and technology grew out of the work of prior decades, the research of today is paving the way for the next generation.
This outlook is reflected in how he interacts with those around him. “I want to have a positive impact on colleagues and students by challenging them, helping them learn, or empowering them to achieve their goals,” says Bo. “Each person’s impact on the world is multiplied by all the people they have an impact on, and as educators this true for us most of all.”
Bo recalls that his interest in computer science started in high school when he explored programming for fun. From there, he went on to receive his B.S. in Mathematics and Computer Science from Duke University and his Ph.D. in Computer Science from Harvard University.
Bo is also an avid runner, and competes for the Boston Athletic Association. He has competed and placed highly in recent 5ks, 10ks, and marathons, and even qualified for the US Olympic Trials marathon in 2016. Bo is currently seeking tenure-track positions in computer science and related disciplines.
To find out more about Bo’s work, check out his website or some of his recent papers: Informational Substitutes, Multi-Observation Elicitation, and Accuracy First: Selecting a Differential Privacy Level for Accuracy-Constrained ERM.