Software algorithms are changing how people work in an ever-growing number of fields. In this study, we explored the impact of this algorithmic, data-driven management on human workers and work practices in the context of Uber and Lyft. I worked on this project as a part of the Social Computing Lab at Carnegie Mellon University.
Our team of three undergraduate CMU students and a post-doctorate student started the project by doing research on previous literature involving the impact of technology in the workplace, management of human workers both by humans and algorithms, trust and perceptions of fairness in relation to technology and cooperation with technology.
The next step in our study was to interview Uber and Lyft drivers and passengers. The goal of our interviews with drivers was to get a better understanding of drivers' experiences working with an algorithm that automatically assigns them passengers and calculates the price of each ride. We particularly wanted to understand drivers' motivation to drive for ride-sharing services, drivers' understanding of how the algorithm works and if the algorithm ever did not work so well with drivers. We also spoke to Uber and Lyft passengers so as to confirm or disconfirm drivers’ perceptions of passengers’ use of services, in particular, how they rate drivers and their attitudes and behaviors around surge pricing.
During this step, I helped design a discussion guide and interview drivers. However, I was not able to contribute to many interviews because I was working outside of Pittsburgh at the time.
The last step in our study was analysis. We used Dedoose, an application for mixed methods research, to analyze interviews and come up with themes. We also used group brainstorming and theming sessions to come up with findings.
Driver motivation to drive for ride-sharing services comes from the flexibility that the system affords in terms of where and when to work, and the low level of commitment that is required by signing up.
Uber and Lyft drivers are encouraged to accept and cooperate with algorithmic assignments. This decision is influenced by the way that assignments were presented to drivers and whether the assignment made sense to the driver.
Drivers used their understanding that assignments are based on proximity to create their work and workaround strategies that helped them maintain control that the automated assignment algorithm functionally did not support.
Drivers benefited from deeper knowledge of the assignment algorithm. Drivers with more knowledge created workarounds to avoid undesirable assignments, whereas those with less knowledge rejected undesirable assignments which lowered their acceptance rating, or unwillingly fulfilled the uneconomical rides.
Acceptance rate threshold systems pushed drivers to accept more rides than they wanted to. This was seen as unfair and ineffective and created negative psychological feelings in drivers.
As drivers worked independently in distributed locations, online driver forums became a primary avenue for the driver socialization and system sense-making.
Our paper was accepted to CHI (Conference on Human Factors in Computing Systems). We made a video for this conference which I helped write and act in.
You can read our paper here.
For part of the study, I was working outside of Pittsburgh as a UX research at LinkedIn. For this reason, I was not able to be counted as an author of the paper.