Presentation in Self-Posted Facial Images Can Expose Sexual Orientation: Implications for Research and Privacy
Bibliographic record
Abstract
Recent research has found that facial recognition algorithms can accurately classify people’s sexual orientations using naturalistic facial images, highlighting a severe risk to privacy. This article tests whether people of different sexual orientations presented themselves distinctively in photographs, and whether these distinctions revealed their sexual orientation. I found significant differences in self-presentation. For example, gay individuals were on average more likely to wear glasses compared to heterosexual individuals in images uploaded to the dating website. Gay men also uploaded brighter images compared to heterosexual men. To further test how some of these differences drove the classification of sexual orientation, I employed image augmentation or modification techniques. To evaluate whether the image background contributed to classifications, I progressively masked images until only a thin border of image background remained in each facial image. I found that even these pixels classified sexual orientations at rates significantly higher than random chance. I also blurred images, and found that merely three numbers representing the brightness of each color channel classified sexual orientations. These findings contribute to psychological research on sexual orientation by highlighting how people chose to present themselves differently on the dating website according to their sexual orientations, and how these distinctions were used by the algorithm to classify sexual orientations. The findings also expose a privacy risk as they suggest that do-it-yourself data-protection strategies, such as masking and blurring, cannot effectively prevent leakage of sexual orientation information. As consumers are not equipped to protect themselves, the burden of privacy protection should be shifted to companies and governments.
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 | 0.155 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.002 | 0.006 |
| Scholarly communication | 0.005 | 0.009 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.008 | 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".