Assessing and Addressing Algorithmic Bias - But Before We Get There

There is a significant gap between awareness and action: While algorithmic bias has become a popular concern, the paper notes that "standard processes and tools for practitioners do not readily exist to assess and address unfair algorithmic and data biases," meaning organizations often lack practical guidance on how to actually implement bias assessments.

The problem is fragmented across disciplines: The paper identifies that "very different communities are working on the topic," which creates a scattered literature that makes it difficult for practitioners to find cohesive solutions. A truly interdisciplinary approach is needed to address algorithmic bias effectively.

Both data and algorithms require scrutiny: The paper emphasizes the need to assess "both intended data and algorithm characteristics and unintended, unfair biases," suggesting that bias can originate from multiple sources—not just the algorithm itself, but also from the data used to train it.

Practical frameworks are still emerging: The authors describe their work as "an early approach that attempts to translate the literature into processes for (production) teams," indicating that actionable, production-ready methodologies for bias assessment are still in development and not yet standardized across the industry.

Implementation challenges are substantial: Beyond theoretical understanding, the paper highlights "a number of challenges encountered in assessing and addressing algorithmic and data bias in practice," suggesting that real-world deployment of bias mitigation is more complex than academic literature alone suggests.