Assumptions About Algorithms' Capacity for Discrimination.

Area of Science:

• Social Psychology
• Human-Computer Interaction
• Algorithmic Bias

Background:

• Algorithms are increasingly used in decision-making processes across various domains.
• Concerns exist regarding algorithms' potential to perpetuate and obscure discrimination.
• Public perception of algorithmic fairness is a critical factor in their adoption.

Purpose of the Study:

• To investigate whether people perceive algorithms as less discriminatory than humans.
• To examine how beliefs about algorithmic accuracy and emotionality influence preferences for algorithmic evaluation.
• To assess the impact of algorithmic training data on user preferences when discrimination is anticipated.

Main Methods:

• Seven empirical studies were conducted to test hypotheses about algorithmic bias perception.
• Participants' preferences for human versus algorithmic evaluation were measured under various conditions.
• Information about algorithmic training data derived from human judgments was manipulated.

Main Results:

• Participants generally assumed algorithms discriminate less than humans, attributing this to higher accuracy and lower emotionality.
• Individuals were more inclined to be evaluated by algorithms when anticipating possible discrimination.
• Awareness of algorithms training on human data did not significantly alter preferences for algorithmic evaluation in anticipated discrimination scenarios.

Conclusions:

• Algorithmic systems are perceived as less discriminatory than human evaluators, fostering greater user comfort.
• This perception may lead to an overreliance on algorithms, potentially overlooking inherent biases.
• Further research is needed to address the nuanced understanding of algorithmic fairness and its societal implications.