CoChat

Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty

Open access

The DOI matches its Crossref record; 228 of the 229 references it cites were found in a registry or an index, and 1 are not checked yet; 41 of the 135 sources its claims cite were checked for support.

Abstract

Algorithmic transparency entails exposing system properties to various stakeholders for purposes that include understanding, improving, and contesting predictions. Until now, most research into algorithmic transparency has predominantly focused on explainability. Explainability attempts to provide reasons for a machine learning model's behavior to stakeholders. However, understanding a model's specific behavior alone might not be enough for stakeholders to gauge whether the model is wrong or lacks sufficient knowledge to solve the task at hand. In this paper, we argue for considering a complementary form of transparency by estimating and communicating the uncertainty associated with model predictions. First, we discuss methods for assessing uncertainty. Then, we characterize how uncertainty can be used to mitigate model unfairness, augment decision-making, and build trustworthy systems. Finally, we outline methods for displaying uncertainty to stakeholders and recommend how to collect information required for incorporating uncertainty into existing ML pipelines. This work constitutes an interdisciplinary review drawn from literature spanning machine learning, visualization/HCI, design, decision-making, and fairness. We aim to encourage researchers and practitioners to measure, communicate, and use uncertainty as a form of transparency.

Verified 2026-11-19 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

This paper argues that estimating and communicating uncertainty should be treated as a complementary form of algorithmic transparency alongside explainability. It reviews methods for measuring uncertainty and discusses how uncertainty can support fairness, decision-making, trust, and stakeholder communication.

Key findings

  • Uncertainty is split into aleatoric uncertainty (irreducible noise/class overlap) and epistemic uncertainty (reducible with more data); calibration is a precursor to using uncertainty as transparency, and deep ensembles are described as one of the best-performing uncertainty quantification approaches for neural networks, retaining calibration under dataset shift.
  • Uncertainty can help address fairness issues: representation bias is epistemic and may be detected via large epistemic uncertainty on validation/deployment data, while feature and label noise are aleatoric and can distort fairness assessments; even a small amount of label noise can greatly impact fairness evaluation.
  • In decision-making, uncertainty supports reject options and decision-theoretic action selection; JDM findings indicate risk is considered together with expected outcome, asymmetrically, and tolerance for uncertainty decreases superlinearly as stakes increase.
  • Communication is difficult: in a 2010 US/Germany survey, 20% could not identify 10% as the biggest disease risk among 1%, 5%, and 10%, and almost 30% could not answer whether 1 in 10, 1 in 100, or 1 in 1000 represents the largest risk; people interpreted the IPCC’s “very likely” as around 60% probability, whereas the guideline specifies greater than 90%. Quantile dot plots and hypothetical outcome plots (HOPs) are reported to improve some probability/uncertainty judgments.

Methods

  • Interdisciplinary literature review, not an empirical study, drawn from machine learning, visualization/HCI, design, decision-making, and fairness.
  • Uses a recurring breast cancer diagnostics scenario to illustrate uncertainty in fairness, decision-making, trust formation, and communication.
  • Synthesizes uncertainty quantification approaches (Bayesian methods, frequentist methods such as ensembles), evaluation metrics (calibration, NLL, Brier score, ECE), and communication methods (categories, numerical ranges, summary statistics, full distributions, visualizations).

Limitations

  • The paper states that its findings should not be treated as conclusive for predicting real-world usability and recommends testing with real users in concrete tasks and settings.
  • The paper states that empirical understanding of how decision-makers use aleatoric versus epistemic uncertainty is limited, that JDM literature has mostly focused on discrete outcomes, and that there is not a good understanding of how people perceive uncertainty over continuous outcomes such as error bars.
  • Evident from the text: as a review, it provides no primary empirical validation or systematic review protocol; several claims are speculative or drawn from other fields and are framed as open questions for future research.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3461702.3462571

Cites

228 of 229 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Top: The same prediction (no cancer) is made in two different hypothetical cancer diagnosis scenarios. However, our model is much more confident in the first. This is reflected in the predictive entropy for each case (dashed red line denotes the maximum entropy for 3-way classification). Bottom: In a regression task, a predictive distribution contains rich information about our model's predictions (modes, tails, etc). We summarise predictive distributions with means and error-bars (here standard deviations).
Figure 1 Top: The same prediction (no cancer) is made in two different hypothetical cancer diagnosis scenarios. However, our model is much more confident in the first. This is reflected in the predictive entropy for each case (dashed red line denotes the maximum entropy for 3-way classification). Bottom: In a regression task, a predictive distribution contains rich information about our model's predictions (modes, tails, etc). We summarise predictive distributions with means and error-bars (here standard deviations).CC-BY · From the PDF
Figure 2 Uncertainty quantification and evaluation. Left three plots: A 15-element deep ensemble provides increased model (epistemic uncertainty) in data sparse regions. A homoscedastic Gaussian noise model provides aleatoric uncertainty matching the noise in the data. Both combine to produce a predictive distribution. Right: calibration plot for classification task. Each bar corresponds to a bin in which predictions are grouped. Their height corresponds to their proportion of correct predictions.
Figure 2 Uncertainty quantification and evaluation. Left three plots: A 15-element deep ensemble provides increased model (epistemic uncertainty) in data sparse regions. A homoscedastic Gaussian noise model provides aleatoric uncertainty matching the noise in the data. Both combine to produce a predictive distribution. Right: calibration plot for classification task. Each bar corresponds to a bin in which predictions are grouped. Their height corresponds to their proportion of correct predictions.CC-BY · From the PDF
Image 3
CC-BY · From the PDF
(b) Quantile dot plot, which can be used to show the uncertainty around the predicted likelihood of a patient having breast cancer.
(b) Quantile dot plot, which can be used to show the uncertainty around the predicted likelihood of a patient having breast cancer.CC-BY · From the PDF
(c) Cone of uncertainty, showing the uncertainty around the predicted path of the center of a hurricane. Taken from [93].
(c) Cone of uncertainty, showing the uncertainty around the predicted path of the center of a hurricane. Taken from [93].CC-BY · From the PDF
(d) Fanchart, which can be used to show how the predicted crime rate of a city evolves over time.
(d) Fanchart, which can be used to show how the predicted crime rate of a city evolves over time.CC-BY · From the PDF

Verification

Claims the paper makes

28 Valid · 10 Not valid · 60 Not checked · 104 Own results and interpretations

74 more claims · 11 with their cited sources checked

Create a free account to see all 202 claims and to verify the 24 references still open.

See all claims — free account
  • It may state two separate claims as one. 40
  • Probably not a substantial claim. 105

Create a free account to see these claims.

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?

Self-check

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

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 229 references

Research works printed but not linked: 12

  1. Emily Badger , Claire Cain Miller , Adam Pearce, and Kevin Quealy. 2018 . Income Mobility Charts for Girls, Asian- Americans and Other Groups . Or Make YourOwn. (2018). Emily Badger, Claire Cain Miller, Adam Pearce, and Kevin Quealy. 2018. Income Mobility Charts for Girls, Asian-Americans and Other Groups. Or Make YourOwn. (2018).
  2. J Martin Bland and Douglas G Altman . 1998. Bayesians and frequentists. BMJ317, 7166 ( 1998 ), 1151--1160. https://doi.org/10.1136/bmj.317.7166.1151arXiv:https://www.bmj.com/content/317/7166/1151.1.full.pdf 10.1136/bmj.317.7166.1151arXiv:https: J Martin Bland and Douglas G Altman. 1998. Bayesians and frequentists. BMJ317, 7166 (1998), 1151--1160. https://doi.org/10.1136/bmj.317.7166.1151arXiv:https://www.bmj.com/content/317/7166/1151.1.full.pdf · doi:10.1136/bmj.317.7166.1151arxiv:https://www.bmj.com/content/317/7166/1151.1.full.pdf
  3. Shelly Chaiken . 1999. The heuristic-systematic. Dual-process theories social psychology 73 ( 1999 ). Shelly Chaiken. 1999. The heuristic-systematic. Dual-process theories social psychology 73 (1999).
  4. Marvin S Cohen , Raja Parasuraman , and Jared T Freeman . 1998 . Trust in decision aids: A model and its training implications . In Proc. Command and Control Research and Technology Symp. Citeseer. Marvin S Cohen, Raja Parasuraman, and Jared T Freeman. 1998. Trust in decision aids: A model and its training implications. In Proc. Command and Control Research and Technology Symp. Citeseer.
  5. Yarin Gal . 2016. Uncertainty in deep learning . University of Cambridge 1, 3( 2016 ). Yarin Gal. 2016. Uncertainty in deep learning. University of Cambridge 1, 3(2016).
  6. Malik Sajjad Ahmed Nadeem , Jean-Daniel Zucker , and Blaise Hanczar . 2009 . Accuracy-Rejection Curves (ARCs) for Comparing Classification Methods with a Reject Option . In Proceedings of the third International Workshop on Machine Learning in Systems Biology (Proceedings of Machine Learning Research), Saso Dzeroski, Pierre Guerts, and Juho Rousu (Eds.) , Vol. 8 . PMLR, Ljubljana, Slovenia, 65--81. http://proceedings.mlr.press/v8/nadeem10a.html Malik Sajjad Ahmed Nadeem, Jean-Daniel Zucker, and Blaise Hanczar. 2009.Accuracy-Rejection Curves (ARCs) for Comparing Classification Methods with a Reject Option. In Proceedings of the third International Workshop on Machine Learning in Systems Biology (Proceedings of Machine Learning Research), Saso Dzeroski, Pierre Guerts, and Juho Rousu (Eds.), Vol. 8. PMLR, Ljubljana, Slovenia, 65--81. http://proceedings.mlr.press/v8/nadeem10a.html
  7. 10.5555/2888116.2888120 · doi:10.5555/2888116.2888120
  8. Jennifer Preece , Helen Sharp , and Yvonne Rogers . 2015. Interaction design: beyond human-computer interaction . John Wiley & Sons . Jennifer Preece, Helen Sharp, and Yvonne Rogers. 2015. Interaction design: beyond human-computer interaction. John Wiley & Sons.
  9. S Shyam Sundar . 2008. The MAIN model : A heuristic approach to understanding technology effects on credibility . MacArthur Foundation Digital Media andLearning Initiative . S Shyam Sundar. 2008.The MAIN model: A heuristic approach to understanding technology effects on credibility. MacArthur Foundation Digital Media andLearning Initiative.
  10. 10.5555/3305890.3306024 · doi:10.5555/3305890.3306024
  11. Anne Marthe van der Bles , Sander van der Linden , Alexandra L. J. Freeman , James Mitchell , Ana B. Galvao , Lisa Zaval , and David J. Spiegelhalter . 2019 . Communicating Uncertainty about Facts , Numbers and Science. 6 , 5 (2019), 181870 . https://doi.org/10/gf2g9j Anne Marthe van der Bles, Sander van der Linden, Alexandra L. J. Freeman,James Mitchell, Ana B. Galvao, Lisa Zaval, and David J. Spiegelhalter. 2019. Communicating Uncertainty about Facts, Numbers and Science. 6, 5 (2019), 181870. https://doi.org/10/gf2g9j
  12. Claus O. Wilke . 2019. Fundamentals of Data Visualization: A Primer on Making Informative and Compelling Figures ( 1 st edition ed.). O'Reilly Media . Claus O. Wilke. 2019. Fundamentals of Data Visualization: A Primer on Making Informative and Compelling Figures(1st edition ed.). O'Reilly Media.

Cited by

See all 300

Metadata

Authors
Bhatt, U., Antorán, J., Zhang, Y., Liao, Q. V., Sattigeri, P., Fogliato, R., Melançon, G., Krishnan, R., Stanley, J., Tickoo, O., Nachman, L., Chunara, R., Srikumar, M., Weller, A., & Xiang, A.
Affiliations
  • University of Cambridge (GB)
  • IBM (United States) (US)
  • Carnegie Mellon University (US)
  • Concordia University (CA)
  • Intel (United States) (US)
  • New York University (US)
  • Partnership on AI, San Francisco, CA, USA
  • Sony AI, Seattle, WA, USA
Venue
Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society
Publisher
ACM
Published
2021-07-21
Pages
401-413
ISSN
3065-8365 (ISSN-L)
OpenAlex
W3183398589
Cited by
300
References
116
Citations by year
2026: 59 · 2025: 62 · 2024: 65 · 2023: 46 · 2022: 48 · 2021: 18
FWCI
16.64
Type
conference paper
Language
en
Open access
Locations
  • Proceedings of the AAAI/ACM Conference on AI Ethics and Society (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • Cambridge University Engineering Department Publications Database (submittedVersion)
  • arXiv (Cornell University)
Funders
  • Microsoft Research
  • Northwestern University
  • Alan Turing Institute (EP/N510129/1., TU/B/000074, N510129)
  • DeepMind
  • Leverhulme Trust
  • Engineering and Physical Sciences Research Council (N510129, EP/N510129/, EP/N510129/1, TU/B/000074, 2275741, EPSRC grant EP/N510129/1, EP/V025279/1)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
Keywords
algorithmic transparency, uncertainty estimation, uncertainty communication, explainability, model fairness, trustworthy systems, decision-making, uncertainty visualization, human-computer interaction, machine learning pipelines, stakeholder awareness

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-24, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-24 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty. https://app.cochat.ai/papers/10.1145/3461702.3462571
The paper
Bhatt, U., Antorán, J., Zhang, Y., Liao, Q. V., Sattigeri, P., Fogliato, R., Melançon, G., Krishnan, R., Stanley, J., Tickoo, O., Nachman, L., Chunara, R., Srikumar, M., Weller, A., & Xiang, A. (2021). Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty. Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society. https://doi.org/10.1145/3461702.3462571

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.