What’s the Appeal? Perceptions of Review Processes for Algorithmic Decisions
Abstract
If you were significantly impacted by an algorithmic decision, how would you want the decision to be reviewed? In this study, we explore perceptions of review processes for algorithmic decisions that differ across three dimensions: the reviewer, how the review is conducted, and how long the review takes. Using a choice-based conjoint analysis we find that people prefer review processes that provide for human review, the ability to participate in the review process, and a timely outcome. Using a survey, we find that people also see human review that provides for participation to be the fairest review process. Our qualitative analysis indicates that the fairest review process provides the greatest likelihood of a favourable outcome, an opportunity for the decision subject and their situation to be fully and accurately understood, human involvement, and dignity. These findings have implications for the design of contestation procedures and also the design of algorithmic decision-making processes.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
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?
Including humans in algorithmic decision-making processes costs time.
“Including humans in algorithmic decision-making processes costs time [20,29,68]”
From Disentangling Fairness Perceptions in Algorithmic Decision-Making: the Effects of Explanations, Human Oversight, and Contestability (2023) · cites this paper as [68]
Malicious actors might use explanations to manipulate decision subjects by conveying untruthful levels of fairness.
“or to manipulate decision subjects by conveying untruthful levels of “fairness” [68].”
From Disentangling Fairness Perceptions in Algorithmic Decision-Making: the Effects of Explanations, Human Oversight, and Contestability (2023) · cites this paper as [68]
Participants considered that humans would treat decision subjects with dignity and compassion.
“and that humans would treat decision subjects with dignity and compassion (e.g., Q.31; also in line with previous research [17,68,94]).”
From Disentangling Fairness Perceptions in Algorithmic Decision-Making: the Effects of Explanations, Human Oversight, and Contestability (2023) · cites this paper as [68]
Participants considered that discussing the decision with humans would potentially lead to a change in outcome for cases that were close to the decision boundary.
“Our participants considered that discussing the decision with humans would potentially lead to a change in outcome for cases that were close to the decision boundary (e.g., Q.27, Q.28; in line with earlier work [36,68])”
From Disentangling Fairness Perceptions in Algorithmic Decision-Making: the Effects of Explanations, Human Oversight, and Contestability (2023) · cites this paper as [68]
The citation lists haven’t been fetched yet
References 57 · Cited by 27
- Authors
- Lyons, H., Wijenayake, S., Miller, T., & Velloso, E.
- Affiliations
- The University of Melbourne (AU)
- The University of Sydney (AU)
- Venue
- CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2022-04-29
- Pages
- 1-15
- OpenAlex
- W4225089167
- Cited by
- 27
- References
- 57
- Citations by year
- 2026: 4 · 2025: 8 · 2024: 8 · 2023: 6 · 2022: 1
- FWCI
- 5.71
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- CHI Conference on Human Factors in Computing Systems (publishedVersion)
- Minerva Access (University of Melbourne) (submittedVersion)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Social Sciences › Psychology › Social Psychology › Human-Automation Interaction and Safety
- Keywords
- human review, algorithmic decision-making, review timeliness, user perception, algorithmic transparency
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
