Rolling in the deep of cognitive and AI biases

Human cognitive biases directly shape AI bias throughout the entire system lifecycle: The paper argues that AI fairness cannot be understood through computational methods alone, as human cognitive biases become "core entities" that influence how AI systems are designed, developed, and deployed in sensitive domains like healthcare, finance, and law enforcement.

Current fairness approaches are incomplete: While numerous studies focus on Machine Learning pipelines and computational fairness metrics, the paper contends this overlooks "the significant impact of human and societal factors on AI bias," explaining why carefully designed AI systems continue to produce discriminatory outcomes despite technical safeguards.

AI systems are sociotechnical, not purely technical: The research emphasizes that AI fairness must be understood as inseparable from "the conditions in which it is designed, developed and deployed," requiring consideration of broader organizational and social contexts beyond algorithm design.

Hidden pathways connect human heuristics to AI biases: The paper introduces a new mapping that reveals how human cognitive heuristics create "hidden pathways" to AI biases, identifying specific connections and "fairness intensities and inter-dependencies" between human decision-making patterns and algorithmic outcomes.

Interdisciplinary perspective is essential: By drawing on cognitive science definitions and taxonomy of human heuristics, the research demonstrates that responsible AI requires understanding both human psychology and technical systems as interconnected components.