Files
Abstract
Given the many notions of fairness, virtually any algorithm or human can be biased or unfair by some standard. Little is known about people’s preferences for potentially biased humans consulting potentially biased Artificial Intelligence (AI) tools. In two studies (N = 1,610), we found that (a) people prefer biased humans to consult unbiased AI, (b) people prefer unbiased humans to avoid consulting biased AI, and (c) people lean towards wanting biased humans to consult biased AI. Although people prefer not to utilize algorithmic aids when the human decision-maker is unbiased, people do not always react negatively to algorithms. Algorithms can be a more permissible decision aid for potentially biased humans. This work suggests that in real-world conditions where any human or algorithm can be biased by some standard, people may not object to potentially biased experts consulting with potentially biased AI.