Algorithmic bias in hiring and lending perpetuates discrimination

Quality and quantity of direct evidence for or against the claim — RCTs, systematic reviews, natural experiments, large cohort studies.

Amazon's scrapped hiring tool and Bartlett et al.'s FinTech lending disparity study document real cases of algorithmic bias, while Buolamwini & Gebru's facial recognition gaps show the problem is real but design-dependent.

Whether the proposed mechanism is valid and established — does the how make sense, or are there fundamental flaws in the causal logic?

The training-data-encodes-historical-bias mechanism is well-established, but audited/fairness-constrained systems can reduce bias relative to human baselines, meaning the mechanism cuts both ways depending on implementation.

Degree of agreement among domain experts and relevant scientific or policy bodies — depth and quality of consensus, not just majority opinion.

AI fairness researchers are genuinely split between documenting harm cases and demonstrating bias-reduction potential through careful system design — a contested rather than settled area.

Whether findings hold across independent studies, populations, and contexts — resistance to p-hacking and publication bias.

Bias findings replicate in specific documented cases (Amazon, facial recognition) but do not generalize uniformly across all algorithmic systems, some of which show reduced bias versus human decision-making.