The Effect of Algorithm Transparency on Algorithm Utilization

In many settings, the pairing of humans with algorithmic aids that assist in judgment and decision-making tasks is rapidly increasing. Though many factors have been shown to affect people’s willingness to rely on algorithms, a large body of evidence across many domains shows that people tend not to do so, even if it would benefit them. The goal of the current research was to investigate whether algorithm transparency affects people’s willingness to utilize algorithms when making future predictions. A secondary goal was to investigate whether algorithm utilization is also influenced by individual differences. Building upon prior research, I developed a theoretical model that accounts for algorithm transparency as well as individual differences in self-enhancement and numeracy. To test this model, I adopted the judge-advisor system for an experiment in which participants completed a series of forecasts in which they received recommendations from an algorithm after being randomly assigned to one of three transparency conditions. The results provide empirical support for the model, suggesting that the extent to which people utilize an algorithm when making forecasts depends on both the transparency of the algorithm and the qualities of the person. Participants who were most confident in their own accuracy, but lower in numeracy, utilized the algorithm the least. This occurred when the algorithm was most transparent. The findings have important theoretical implications for our understanding of algorithm transparency. A primary aspect of this contribution is that it conceptualizes transparency as multidimensional and rooted in perception, and therefore, not strictly a quality of an algorithm. This suggests that even when interacting with the same algorithm of equal transparency, individuals hold different perceptions about how transparent the algorithm is.

Author

• Willford, Jonathan

Language

• en

Keyword

• Forecasting
• Individual Differences
• Decision-making
• Algorithms
• Transparency

Date created

• 2021

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Organizational Sciences

Degree

• Ph.D.

Advisor

• Behrend, Tara S
• Costanza, David P

Committee Member(s)

• Dodge, Tonya
• Olsen, Nils
• Medsker, Larry
• Watkins, Ryan
• Debebe, Gelaye

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/k06988438