Adgorithm
Final Project, User Centered Research & Evaluation (UCRE)
User research and low-fidelity prototype, harnesses everyday users to audit for algorithm bias
Our solution to addressing the issue of implicit ad bias is to increase awareness and interest by gamifying the ad auditing experience. In this way we will expose users to a low commitment, daily game that raises awareness and provides data to companies such as Instagram that use ad algorithms.
Role: ideator, interviewer, principal prototyper
Tools: Miro, Figma
Team members: Sophia Witt, Angela Oh, Megan Yim, Lucia Fang
Duration: 1 semester
Problem Space
Recent events such as the Twitter Cropping Algorithm incident and certain Tiktok filters that are blatantly racist have prompted people to look into algorithm bias and how to combat this issue. However, in the case of ad algorithms, bias is hard to detect since the most prevalent example of biased ads are ads that are directed towards some groups and not others, making them hard for either group to detect as biased.
Goal
Our goal was to find the most effective way to harness everyday users to audit for harmful algorithmic biases.
Contextual Inquiry
From our initial contextual inquiry, I learned that many users understand Instagram's ad algorithm to some extent. In my interviews I noticed several times that people would explain to me why they got a certain type of ad, or even who they thought Instagram's algorithm thought they were. Keeping this in mind, our team set out to find a way to educate users about the harms of ad algorithm bias and to harness these everyday users to audit for these harmful algorithmic biases. I predicted that the fact that many users already have an understanding of ad algorithms this would make our job easier.
Affinity Clustering
From our directed storytelling and semi-structured interviews, we were able to gain several insights including the fact that many users found ads beneficial, since they enjoyed ads for new items they were interested in, and many people felt they didn't experience ad bias, or at least harmful ad bias. This led our group to think that we first had to find a way to raise awareness about the potential harm algorithmic bias can cause as well as educate users on ad bias so that they can better detect it.
Survey
After sending out a survey, we found that many users notice ad bias, but don't think that it is harmful and so don't feel the need to report ads. This confirmed that we needed to find a way to educate users on ad bias so that they can better detect it.
Speed Dating
I made storyboards focused on transparency for the speed dating activity, since I thought that if users had more information, they would be more likely to spot harmful algorithmic bias and report ads. This was mostly confirmed in the speed dating activity.
Initial Prototype (Low Fidelity)
For our initial prototype, I made a very low-fi prototype on Figma where users would type in their answer in free-response form (this was simulated by a click in the prototype) and then click submit to see if they got it right. After they got the right answer, there was a page explaining algorithmic bias, and then a page showing the user's streak with an option to share with their friends.
Final Prototype (Low Fidelity)
From our prototype testing we learned that free-response left too many options to the user, and that it would be really difficult for them to get the right answer. This is why for our final prototype, I modified the original prototype to make things such as gender multiple choice, and for age ranges I added drop-down menus for the user to select from.
Final Thoughts
This was my first course in user research, and I learned so many different methods and tools for getting to know more about users' motivations, behaviors, and needs. I found that I really enjoy the process of trying to get the most holistic understanding as possible of users in the context of the problem space, and considering all the nuances of the problem we are trying to solve. In particular, I have found what I learned about conducting user interviews useful in other courses where I have to conduct interviews with people, particularly in constructing the interview questions and determining the order of the questions (general to specific).