CoChat
Sign Up
Cited by 35Ethics and Social Impacts of AIExplainable Artificial Intelligence (XAI)Artificial Intelligence in Law

The four-fifths rule is not disparate impact: a woeful tale of epistemic trespassing in algorithmic fairness

Open access
Watkins, Elizabeth Anne, Chen, Jiahao·2024·The 2024 ACM Conference on Fairness Accountability and Transparency
DOI 10.1145/3630106.3658938
Open this paper in CoChat — free accountOpen PDF

Abstract

Computer scientists are trained in the art of creating abstractions that simplify and generalize. However, a premature abstraction that omits crucial contextual details creates the risk of epistemic trespassing, by falsely asserting its relevance into other contexts. We study how the field of responsible AI has created an imperfect synecdoche by abstracting the four-fifths rule (a.k.a. the <?TeX $\nicefrac {4}{5}$?> Math 1 rule or 80% rule), a single part of disparate impact discrimination law, into the disparate impact metric. This metric incorrectly introduces a new deontic nuance and new potentials for ethical harms that were absent in the original <?TeX $\nicefrac {4}{5}$?> Math 2 rule. We also survey how the field has amplified the potential for harm in codifying the <?TeX $\nicefrac {4}{5}$?> Math 3 rule into popular AI fairness software toolkits. The harmful erasure of legal nuances is a wake-up call for computer scientists to self-critically re-evaluate the abstractions they create and use, particularly in the interdisciplinary field of AI ethics.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1145/3630106.3658938

Cites

43 of 44 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • If the less-selected group's selection rate falls below four-fifths (80%) of the baseline then there is a presumption that disparate impact may be occurring and warrants further investigation, typically through a complaint to the EEOC.

    “If the less-selected group's selection rate falls below four-fifths (80%) of the baseline then there is a presumption that disparate impact may be occurring and warrants further investigation, typically through a complaint to the EEOC.”

    From Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability (2024) · cites this paper as [69]

    The appendix’s quotation of the Uniform Guidelines (§1607.4(D); page not specified) says a selection rate below four-fifths of the highest group’s rate is generally regarded as evidence of adverse impact. Section 2.3 describes the rule as a trigger for further regulatory scrutiny. Neither passage says that investigation typically proceeds through a complaint to the EEOC. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

  • The four-fifths rule is not genuinely a rule or law; it is best described as a rule of thumb or convention that is ingrained in hiring practices, and employers and developers are highly incentivized to reduce potential scrutiny by targeting that threshold.

    “The four-fifths rule is not genuinely a rule or law; it is best described as a rule of thumb or convention that is ingrained in hiring practices, and employers and developers are highly incentivized to reduce potential scrutiny by targeting that threshold.”

    From Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability (2024) · cites this paper as [69]

    Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Experimental — accuracy not yet measured

  • The four-fifths rule establishes a baseline against which impermissible discriminatory outcomes can be identified by comparing relative rates of selection between demographic groups.

    “The four-fifths rule establishes a baseline against which impermissible discriminatory outcomes (aka 'disparate impact') can be identified by comparing relative rates of selection between demographic groups.”

    From Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability (2024) · cites this paper as [69]

    Appendix A.2 describes comparing demographic groups’ selection rates against the highest rate, so the rule does provide a screening threshold. But it calls the rule a rule of thumb, not a legal definition; the section on selecting a relevant comparison population says it is neither necessary nor sufficient for a court finding of disparate impact. The claim overstates what the comparison can establish by saying it identifies impermissible discriminatory outcomes. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

  • The four-fifths rule is a general guideline that an impact ratio greater than 0.8 is considered de facto evidence (but not de jure evidence) of non-discriminatory outcomes.

    “The 'four-fifths rule' is a general guideline that an impact ratio greater than 0.8 is considered de facto evidence (but not de jure evidence) of non-discriminatory outcomes.”

    From Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability (2024) · cites this paper as [69]

    The quoted regulation in Appendix A says that federal agencies generally do not regard an impact ratio above 0.8 as evidence of adverse impact, and Section 2.3 describes the rule as a practical screening guideline rather than a legal definition. But not treating a ratio as evidence of adverse impact is not affirmative evidence of non-discriminatory outcomes. Section 2.3 also says the rule has not generally been extended beyond employment decisions. (LLM, conf=0.87)

    Experimental — accuracy not yet measured

  • As critics have noted, the four-fifths rule is a fundamentally arbitrary threshold not supported by robust empirical evidence.

    “As critics have noted [69], the four-fifths rule is a fundamentally arbitrary threshold not supported by robust empirical evidence.”

    From Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability (2024) · cites this paper as [69]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 12

  1. Michelle Bao, Angela Zhou, Samantha A Zottola, Brian Brubach, Sarah Desmarais, Aaron Seth Horowitz, Kristian Lum, and Suresh Venkatasubramanian. 2021. It's COMPASlicated: The Messy Relationship between RAI Datasets and Algorithmic Fairness Benchmarks. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1) (NeurIPS '21). OpenReview, OpenReview.net, 18 pages. https://openreview.net/forum?id=qeM58whnpXM
  2. Felice Cardone. 2021. Games, Full Abstraction and Full Completeness. In The Stanford Encyclopedia of Philosophy (Spring 2021 ed.), Edward N. Zalta (Ed.). Metaphysics Research Lab, Stanford University, Stanford, CA.
  3. Samuel Deng and Achille Varzi. 2019. Methodological blind spots in machine learning fairness: Lessons from the philosophy of science and computer science. In Workshop on Human-Centric Machine Learning at the 33rd Conference on Neural Information Processing Systems (NeurIPS 2019). arXiv, Vancouver, Canada, 5 pages. arXiv:1910.14210 · arXiv
  4. Paul London. 1978. A Conversation with Eleanor Holmes Norton. Employee Relations Law Journal 3 (1978), 314-326.
  5. David Manheim and Scott Garrabrant. 2018. Categorizing variants of Goodhart's Law. arXiv:1803.04585 · arXiv
  6. Donald Martin Jr, Vinodkumar Prabhakaran, Jill Kuhlberg, Andrew Smart, and William S Isaac. 2020. Extending the machine learning abstraction boundary: A Complex systems approach to incorporate societal context. arXiv:2006.09663 · arXiv
  7. Nick Merrill, Michael Madaio, and Richmond Wong. 2022. Seeking Like a Toolkit: How Toolkits Envision the Work of AI Ethics. arXiv:2202.08792 · arXiv
  8. Amir Radfar, Seyed Ahmad Ahmadi Asgharzadeh, Fernando Quesada, and Irina Filip. 2018. Challenges and perspectives of child labor. Industrial psychiatry journal 27, 1 (Jan-Jun 2018), 17-20.
  9. Benjamin I. Sachs. 2007-2008. Employment Law as Labor Law. Cardozo Law Re- view 29 (2007-2008), 2685-2748. https://heinonline.org/HOL/P?h=hein.journals/ cdozo29&i=2707
  10. Pedro Saleiro, Benedict Kuester, Loren Hinkson, Jesse London, Abby Stevens, Ari Anisfeld, Kit T. Rodolfa, and Rayid Ghani. 2019. Aequitas: A Bias and Fairness Audit Toolkit. arXiv:1811.05577 [cs.LG] · arXiv
  11. Edouard J. Simon, Monique Janneck, and Dorina Gumm. 2006. Understanding Socio-Technical Change: Towards a Multidisciplinary Approach. In Social In- formatics: An Information Society for all? In Remembrance of Rob Kling, Jacques Berleur, Markku I. Nurminen, and John Impagliazzo (Eds.). Springer US, Boston, MA, 469-479.
  12. Alice Xiang and Inioluwa Deborah Raji. 2019. On the legal compatibility of fairness definitions. In Workshop on Human-Centric Machine Learning at the 33rd Conference on Neural Information Processing Systems (NeurIPS '19). arXiv, Vancouver, Canada, 6 pages. arXiv:1912.00761 · arXiv

Cited by

Sorted by citations

Related

Authors
Watkins, E. A., & Chen, J.
Affiliations
  • Intel (United States) (US)
  • Responsible AI LLC, United States of America
ORCID
Venue
The 2024 ACM Conference on Fairness Accountability and Transparency
Publisher
ACM
Published
2024-06-03
Pages
764-775
DOI
10.1145/3630106.3658938
OpenAlex
W4399364234
Cited by
25
References
42
Citations by year
2026: 4 · 2025: 7 · 2024: 6 · 2023: 4 · 2022: 4
FWCI
7.75
Type
conference paper
Language
en
Open access
Licence
www.acm.org/publications/policies/copyright_policy#Background (vor)
Locations
  • ACM Conference on Fairness, Accountability, and Transparency (FAccT) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Social Sciences › Political Science and International Relations › Artificial Intelligence in Law
Keywords
indirect discrimination, algorithmic fairness, responsible AI, legal abstraction, algorithmic discrimination, AI ethics, fairness metrics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review