Cross-model Fairness: Empirical Study of Fairness and Ethics Under Model Multiplicity

This repository holds Jupyter Notebooks to reproduce experimental results and plots from the Cross-model Fairness: Empirical Study of Fairness and Ethics Under Model Multiplicity paper published in the ACM Journal on Responsible Computing.

Code

The experiments have been executed with Python 3.8.2. The packages and their versions are captured by the requirements.txt file (pip install -r requirements.txt). Accessing OpenML requires an API token. This can either be set up directly in the openml.ipynb Jupyter Notebook or placed in a BASH variable named OPENML_APIKEY (export OPENML_APIKEY=myOpenmlAPIkey). The following notebooks are provided:

• toy_plots.ipynb -- generates toy example plots.
• openml.ipynb -- fetches data from OpenML.
• plot_profile.ipynb -- plots fairness profiles.
• ambiguity.ipynb -- computes and graphs ambiguity and discrepancy.
• fair_model.ipynb -- analyses performance of individually fair models.

Note that downloading experiments from OpenML may take a long time given that they are made available as a large number of relatively small files (roughly 10GB in total).

Paper

Cross-model Fairness: Empirical Study of Fairness and Ethics Under Model Multiplicity

Abstract:

While data-driven predictive models are a strictly technological construct, they may operate within a social context in which benign engineering choices entail implicit, indirect and unexpected real-life consequences. Fairness of such systems -- pertaining both to individuals and groups -- is one relevant consideration in this space; algorithms can discriminate people across various protected characteristics regardless of whether these properties are included in the data or discernible through proxy variables. To date, this notion has predominantly been studied for a fixed model, often under different classification thresholds, striving to identify and eradicate undesirable, discriminative and possibly unlawful aspects of its operation. Here, we backtrack on this fixed model assumption to propose and explore a novel definition of cross-model fairness where individuals can be harmed when one predictor is chosen ad hoc from a group of equally well performing models, i.e., in view of utility-based model multiplicity. Since a person may be classified differently across models that are otherwise considered equivalent, this individual could argue for a predictor granting them the most favourable outcome, employing which may have adverse effects on other people. We introduce this scenario with a two-dimensional example and linear classification; then, we present a comprehensive empirical study based on real-life predictive models and data sets that are popular with the algorithmic fairness community; finally, we investigate analytical properties of cross-model fairness and its ramifications in a broader context. Our findings suggest that such unfairness can be readily found in real life and it may be difficult to mitigate by technical means alone as doing so is likely to degrade predictive performance.

Keywords: Rashomon effect, epistemic uncertainty, robustness, machine learning, artificial intelligence.

Highlights:

• Cross-model fairness guarantees that a person receives the same prediction for a collection of classifiers with identical or comparable predictive performance, i.e., under utility-based model multiplicity.
• When at least one prediction output by multiple equivalent models for a single individual is perceived as favourable, the person might argue for the precedence of this outcome.
• Cross-model fairness is consistent with the Blackstone's ratio and "presumption of innocence" -- lack of convincing evidence ought to warrant the most favourable treatment.
• Granting each person the most favourable decision afforded by a collection of equivalent models may degrade the overall predictive performance on the underlying task, especially when the employed family of models is highly expressive.
• A possible solution is to limit the number of admissible predictors by imposing appropriate modelling restrictions that are consistent with the social context and the natural process governing the generation of the underlying data, as well as to employ a justified prediction aggregation strategy across classifiers with identical or comparable performance.

Citation:

@article{sokol2024cross,
  title={Cross-model Fairness:
         {Empirical} Study of Fairness and Ethics Under Model Multiplicity},
  author={Sokol, Kacper and Kull, Meelis and Chan, Jeffrey and Salim, Flora},
  journal={ACM Journal on Responsible Computing},
  publisher={ACM New York, NY},
  doi={10.1145/3677173},
  year={2024}
}