Algorithm aversion is when an AI is discriminated against. There are a variety of approaches for studying this phenomena which clearly exists.
Algorithm aversion
Aversion is assumed to be an irrational and largely unconscious preference against something or someone. Algorithm aversion is an implicit prejudice against AIs in the widest sense of the word AI (namely non-human computation).
The key to reducing algorithm aversion is enabling the interaction with algorithms under conditions which reveal its applicability and effectiveness. There are several dimensions to this.
Algorithm aversion research
- Comparing discriminatory behavior against AI and humans
- Found that choice congruency explains descrimintary behavior, rather than some essential AI characteristic.
- The Algorithm of Fear - Unpacking Prejudice Against AI
- Human exceptionalism is challenged in recent studies of AI performance, yet public doubt in the models' abilities via long-standing anxieties hinder their adoption.
- Algorithm Aversion - Erroneously Avoid Algorithms after Seeing Them Err
- Humans more quickly lose confidence in an AI model making mistakes, than they do with humans, even when it was demonstrated that the models performed better.
- When Self-Humanization Leads to Algorithm Aversion
- The importance of empathy and the importance of autonomy affect the context in which AI may or may not be welcome. The decision support that is most welcome would be sensitive to the kinds of user interactions supported which would include or enable self-actualization.
- Task-Dependent Algorithm Aversion
- Humans trust algorithms less with tasks perceived as subjective rather than objective. Increasing the perceived subjectivity of the algorithm and the perceived objectivity of the task can enable better acceptance and fit.
- Algorithm Aversion: Who Trusts Algorithms to Make Decisions?
- Age, gender, extraversion, and openness significantly affect algorithm aversion. Transparency (clear understanding and context awareness) and lowered costs of the algorithm models decrease algorithm aversion.