The Bias-Boomerang: When Anti-Bias Measures Backfire
Bibliographic record
Abstract
Algorithmic bias refers to systematic and repeatable errors within a computer system that generate unfair outcomes, often favoring one group of users over others. Bias can enter through historically skewed training data, self-reinforcing feedback loops that perpetuate prior patterns, or design choices made during the development of the algorithm itself. Existing literature on algorithmic bias has discussed aspects such as how bias materializes in algorithmic outputs (Obermeyer et al., 2019) and the underlying epistemic opacity of the algorithms hindering experts’ capacity to trust a judgment (Lebovitz et al., 2022). What if bias is not an outcome of inattention to vulnerable populations or opacity of algorithms, but emanates from the specific attempts to help the vulnerable populations and to address the opacity of algorithms? Recently, limited research offers some evidence of this possibility of bias backfiring (Yan et al., 2024). Such impact can undermine trust in AI (Artificial Intelligence) systems, especially when fairness claims are made without transparency about trade-offs or unintended consequences. In order to explore this phenomenon, I intend to undertake an empirical study within a financial institution that actively engages in the development and use of algorithmic systems. The setting is relevant since financial services are high-stakes contexts, where algorithmic decisions can significantly impact individuals’ access to credit, insurance, and other financial products. The study will seek to answer two research questions: First, what specific guardrails and oversight mechanisms does the organization implement to assess the potential consequences of a bias mitigation strategy before its deployment. Second, once implemented, how does the organization evaluate the effectiveness of these checks in practice, and what mechanisms are employed to ensure that such interventions do not generate adverse outcomes over time. By examining these questions, the study aims to contribute to the growing body of research on fairness in algorithms, by providing an empirical account of how practitioners navigate the risk of bias-mitigation measures inadvertently backfiring.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.039 | 0.224 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.007 | 0.004 |
| Science and technology studies | 0.005 | 0.028 |
| Scholarly communication | 0.009 | 0.019 |
| Open science | 0.004 | 0.012 |
| Research integrity | 0.006 | 0.007 |
| Insufficient payload (model declined to judge) | 0.014 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".