Human-in-the-Loop Oversight of AI is Compromised by Political Preferences
Year: 2024
Type: preprint
Abstract: Experiment in which humans are used to provide feedback to an algorithm used to give recommendations about which social support applications, submitted by immigrant and locals, are approved. The algorithm is varied to be neutral, biased towards immigrants, or biased towards locals. The aim is to test whether these humans in the loop create rather than mitigate bias in algorithms based on their political orientation (democrat voters favoring benefits to immigrants more than republican voters; republican voters favoring benefits to locals more than republican voters).
Institution: Vrije Universiteit Amsterdam
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7105734107
