Algorithmic Face-ism: Uncovering and Mitigating Algorithmic Bias in Facial Recognition Systems
Bibliographic record
Abstract
How do some of the most advanced machine learning facial recognition algorithms make important decisions, such as whom to hire or who is considered a leader? Existing research suggests advances in machine learning methods can use facial features in an image—facial morphology—to accurately and objectively predict answers to such questions. We show, however, that even after implementing state-of-the-art models, decision-based facial recognition algorithms are not as objective as previously claimed. Unpacking the “black box” of an existing facial recognition algorithm revealed the algorithm did not rely on facial morphology to make decisions. Instead, when covariates such as attractiveness were accounted for, the algorithm relied mostly on “leftover” transient features, such as clothing, hairstyle, or background lighting for decision-making. We identify the specific stages—sampling, preprocessing, model implementation, and model functioning stage— in which algorithmic focus bias and interpretation bias are likely to arise in facial recognition algorithms. These results suggest that decision-based facial recognition algorithms are biased in ways that researchers have overlooked, with troubling implications for their use by governments, organizations, and researchers. We introduce the concept of “algorithmic face-ism,” in which (1) machine learning algorithms unfairly express an inherent preference for specific facial morphologies, and (2) researchers mistakenly attribute behavioral predictions to facial morphologies. This paper thus demonstrates how leading decision-based facial recognition systems are biased and how previously taken-for-granted factors contribute to this pattern of bias. We conclude by discussing how bias can be mitigated in such facial recognition algorithms.
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 | 0.021 | 0.118 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.004 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".