Facilitating secure data sharing between federal agencies.

Challenge

xD (an experimental group at the U.S. Census Bureau) develops emerging technologies, including secure methods of data sharing. My team was exploring Secure Multi-Party Computation (SMPC), a process that would allow data scientists from other agencies to join their own data sets with Census data sets to answer research questions around equity while maintaining the privacy of individuals. Because many data scientists are not familiar with Secure Multi-Party Computation, I was asked to design a user interface that would transform the raw code and process into a trustworthy and easy-to-use platform.

Approach

I first spoke with data scientists at other agencies to gauge their understanding of Secure Multi-Party Computation and interest in joining their data with specific Census data sets. After better understanding their needs and points of confusion, I designed an interface for a simple web application using familiar US Web Design System components. This platform, called SMPC Data Joiner, introduces researchers to the SMPC process, allows them to easily upload data they want to join with Census data, and view the results once they have gone through the appropriate reviews.

Impact

While this project was paused indefinitely due to factors outside of our team's control, I feel confident that the research and design I completed laid a solid foundation for a future team to pick up the work where we left off. I also had the opportunity to advocate for the value of UX and UI in the data science space by demonstrating these designs.

Team

UX/UI Designer
Marina DeFrates

Data Scientist
Tomo Lazovich
Anna Vasylytsya

Project Manager
Samantha Weinstock

Context

Federal government

Timeline

3 weeks
(Spring 2024)

Status

Prototyped

Process

1/ Learning from data scientists

I wanted to gather context for how non-Census data scientists might use the SMPC process and find out key pieces of information, such as:

What kinds of data sets they want to work with
How they access those data sets currently and associated difficulties
How using the SMPC process might be able to make data access and joining easier for them

Key findings + design opportunities

🔍
Interviewees have an interest in working with Census data sets that could be joined using SMPC.

Translate a complicated process into a clear, efficient flow for data joining.

🔍
All interviewees indicated that the current review process is unclear and takes a long time.

Set clear expectations throughout regarding process steps and timeline.

🔍
Some interviewees were familiar with PETs generally but not SMPC specifically and were curious about the security.

Build trust in the SMPC technology and associated processes.

2/ Translating a complicated process into a simple flow

With the help of my team, I was able to break down the process as it exists in Python and design corresponding screens.

Designs

Breaking down the steps

This flow would allow data scientists to select the data set they want to join with, upload their own data, confirm the join and be alerted as each step of is completed without having to deal with the back end.

Setting expectations

Before joining with a Census data set, a banner alerts data scientists about the Disclosure review process they will have to go through and how long they can expect it to take. A "My joins" dashboard also lets data scientists know the status of joins awaiting review and those that are ready to view.

Building trust in the technology

From the Data Joiner homepage, users can learn about how the SMPC process works, using an illustrated example from Google. The more that data scientists understand the process, the more they are able to trust that it is secure.

Reflection

I came to this project knowing very little about Privacy Enhancing Technologies or data science in general. While I was initially worried that my lack of knowledge would be detrimental to the design process, I actually found that it freed me to ask questions until I was able to distill very technically advanced concepts into a format for audiences with varying levels of technical familiarity.