CoChat
Sign Up
Cited by 166Explainable Artificial Intelligence (XAI)Ethics and Social Impacts of AIHuman-Automation Interaction and Safety

Fairness and Explanation in AI-Informed Decision Making

Open access
Angerschmid, Alessa, Zhou, Jianlong, Theuermann, Kevin, Chen, Fang, Holzinger, Andreas·2022·Machine Learning and Knowledge Extraction
DOI 10.3390/make4020026
Open this paper in CoChat — free accountOpen PDF

Abstract

AI-assisted decision-making that impacts individuals raises critical questions about transparency and fairness in artificial intelligence (AI). Much research has highlighted the reciprocal relationships between the transparency/explanation and fairness in AI-assisted decision-making. Thus, considering their impact on user trust or perceived fairness simultaneously benefits responsible use of socio-technical AI systems, but currently receives little attention. In this paper, we investigate the effects of AI explanations and fairness on human-AI trust and perceived fairness, respectively, in specific AI-based decision-making scenarios. A user study simulating AI-assisted decision-making in two health insurance and medical treatment decision-making scenarios provided important insights. Due to the global pandemic and restrictions thereof, the user studies were conducted as online surveys. From the participant’s trust perspective, fairness was found to affect user trust only under the condition of a low fairness level, with the low fairness level reducing user trust. However, adding explanations helped users increase their trust in AI-assisted decision-making. From the perspective of perceived fairness, our work found that low levels of introduced fairness decreased users’ perceptions of fairness, while high levels of introduced fairness increased users’ perceptions of fairness. The addition of explanations definitely increased the perception of fairness. Furthermore, we found that application scenarios influenced trust and perceptions of fairness. The results show that the use of AI explanations and fairness statements in AI applications is complex: we need to consider not only the type of explanations and the degree of fairness introduced, but also the scenarios in which AI-assisted decision-making is used.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper reports an online user study on how AI explanations and introduced fairness jointly affect trust and perceived fairness in AI-assisted health insurance and medical treatment decisions. It finds that low fairness reduces trust and perceived fairness, while explanations generally increase both, with some scenario-dependent differences.

Key findings

  • Low introduced fairness significantly decreased user trust compared with the fairness control condition (t = 3.289, p < 0.001) and with high fairness (t = 5.428, p < 0.000); high fairness did not differ significantly from control (t = 1.953, p < 0.052).
  • Explanations increased trust: trust was significantly lower under the no-explanation control than under feature importance-based explanation (t = 5.656, p < 0.000) and example-based explanation (t = 4.080, p < 0.000), with no significant difference between the two explanation types (t = 1.619, p < 0.107).
  • For perceived fairness, low fairness was significantly lower than high fairness (t = 7.774, p < 0.000) and control (t = 5.477, p < 0.000); high fairness was also higher than control after adjusted alpha (t = 2.281, p < 0.023). Explanations increased perceived fairness versus control (example-based: t = 3.701, p < 0.000; feature importance-based: t = 4.167, p < 0.000), with no difference between explanation types (t = 0.544, p < 0.587).
  • Scenario mattered for explanation effects: in health insurance, feature importance-based explanation produced higher trust than example-based explanation (t = 2.329, p < 0.021), while in medical treatment there was no difference between explanation types (t = 0.000, p = 1); example-based explanations increased perceived fairness in medical treatment but not in health insurance.

Methods

  • Online user study with 25 participants (university students, average age 26, 10 female), simulating AI-assisted decision-making in two scenarios: health insurance payment and medical treatment.
  • Each participant completed 18 tasks: 3 explanation conditions (example-based, feature importance-based, control) × 3 fairness conditions (low, high, control) × 2 scenarios, plus 2 training tasks; task order was randomized.
  • Trust was measured with six 4-point Likert items; perceived fairness with a single 4-point Likert item. Ratings were normalized per subject, then analyzed with two-way ANOVA, one-way ANOVA, and post-hoc t-tests with Bonferroni correction.

Limitations

  • The paper does not state a formal limitations section; it notes that due to the COVID-19 pandemic and restrictions, the user studies were conducted as online surveys.
  • Evident from the text: the sample was small and narrow (25 university students, average age 26, 10 female), limiting generalizability.
  • Evident from the text: fairness was operationalized only through gender as the protected attribute, only two explanation types and two application scenarios were tested, and perceived fairness was measured with a single item.

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

Integrity signals

Registry

Crossref · 10.3390/make4020026

Cites

64 of 65 references verified0 citation errors · 1 not checkable

Statistics

Claims the paper makes

0 of 8 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The addition of explanations increased the perception of fairness.

    “The addition of explanations definitely increased the perception of fairness.”
    Quote in the paperSubstantial claim
  • Findingown data

    Application scenarios influenced trust and perceptions of fairness.

    “Furthermore, we found that application scenarios influenced trust and perceptions of fairness.”
    Quote in the paperSubstantial claim
  • Findingown data

    Adding explanations helped users increase their trust in AI-assisted decision-making.

    “However, adding explanations helped users increase their trust in AI-assisted decision-making.”
    Quote in the paperSubstantial claim
  • Findingnone given

    The application scenarios affect user responses of trust and perception of fairness under different explanation types and introduced fairness levels in AI-informed decision-making.

    “The application scenarios affect user responses of trust and perception of fairness under different explanation types and introduced fairness levels in AI-informed decision-making.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Under feature importance-based explanation, a one-way ANOVA test found significant differences in trust in different introduced fairness levels (F(2, 147) = 4.676, p < 0.011).

    “A one-way ANOVA test found significant differences in trust in different introduced fairness levels under the feature importance-based explanation, F(2, 147) = 4.676, p < 0.011.”
    Quote in the paperSubstantial claim
  • May combine two claims. 17
  • The quote only partly states this claim. 1
  • Classified as probably not a substantial claim. 5

  • May combine two claims.

    Findingown data

    Low levels of introduced fairness decreased users' perceptions of fairness, while high levels of introduced fairness increased users' perceptions of fairness.

    “From the perspective of perceived fairness, our work found that low levels of introduced fairness decreased users' perceptions of fairness, while high levels of introduced fairness increased users' perceptions of fairness.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Fairness affected user trust only under the condition of a low fairness level, with the low fairness level reducing user trust.

    “From the participant's trust perspective, fairness was found to affect user trust only under the condition of a low fairness level, with the low fairness level reducing user trust.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The fairness statement in the scenario affected user trust in AI-informed decision-making only under the low level of fairness condition, where the low-level fairness statement decreased user trust.

    “From the trust's perspective, this study found that the fairness statement in the scenario did affect user trust in AI-information decision-making only under the low level of fairness condition, where the low-level fairness statement decreased user trust in AIinformed decision-making.”
    • Also stated in: § 6. Conclusions · plain
      “A user study by simulating AI-informed decision-making through manipulating AI explanations and fairness levels found that the introduced fairness affected user trust in AI-informed decision-making only under the low level of fairness condition.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Under feature importance-based explanations, similar conclusions were obtained for the perception of fairness as under example-based explanations.

    “Similar conclusions were obtained for the perception of fairness under feature importancebased explanations.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Perception of fairness had a positive effect on trust in an algorithmic decision-making system such as recommendations.

    “Shin's investigations [27,37] showed that perception of fairness had a positive effect on trust in an algorithmic decision-making system such as recommendations.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Shin [27,37]

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    The introduced fairness condition affected user trust in AI-informed decision-making only under the low fairness condition, where introduced fairness decreased user trust.

    “These findings imply that the introduced fairness condition did affect user trust in AI-information decision-making only under the low fairness condition, where introduced fairness decreased user trust in AI-informed decision-making.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The orders of tasks were randomised to avoid any bias introduced.

    “The orders of tasks are randomised to avoid any bias introduced.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Under the example-based explanation condition, the low level of fairness statement significantly decreased the user trust in decision-making, but the high level of fairness statement did not affect user trust.

    “The results showed that, under the example-based explanation condition, the low level of fairness statement significantly decreased the user trust in decision-making, but the high level of fairness statement did not affect user trust.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Explanations helped users increase their trust significantly in AI-informed decision-making, but different explanation types did not show differences in affecting user trust.

    “The results showed that explanations did help users increase their trust significantly in AI-informed decision-making, but different explanation types did not show differences in affecting user trust.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Introduced fairness is positively related to user trust in AI-informed decision-making.

    “Zhou et al. [3] obtained similar conclusions that introduced fairness is positively related to user trust in AI-informed decision-making.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Zhou et al. [3] Zhou, J., Verma, S., Mittal, M., and Chen, F. (2021, January 29–31). Understanding Relations between Perception of Fairness and Trust in Algorithmic Decision Making. Proceedings of the International Conference on Behavioral and Social Computing (BESC 2021), Doha, Qatar. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Introduced fairness is positively related to the perception of fairness, i.e., the high level of introduced fairness resulted in a high level of perception of fairness.

    “It was found that introduced fairness is positively related to the perception of fairness, i.e., the high level of introduced fairness resulted in a high level of perception of fairness.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Zhou et al. [3] Zhou, J., Verma, S., Mittal, M., and Chen, F. (2021, January 29–31). Understanding Relations between Perception of Fairness and Trust in Algorithmic Decision Making. Proceedings of the International Conference on Behavioral and Social Computing (BESC 2021), Doha, Qatar. source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Explainability is indispensable to foster user trust in AI systems, particularly in sensible application domains.

    “Explainability is indispensable to foster user trust in AI systems, particularly in sensible application domains.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Holzinger et al. [47] Holzinger, 2019, Causability and Explainability of Artificial Intelligence in Medicine, Wiley Interdiscip. Rev. Data Min. Knowl. Discov. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Under the high level of introduced fairness, the feature importance-based explanation significantly increased user trust, while the example-based explanation did not.

    “The high level of introduced fairness showed similar effects on user trust as the low level of introduced fairness: the feature importance-based explanation significantly increased user trust, while the example-based explanation did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Under the feature importance-based explanation condition, the low level of fairness statement significantly decreased the user trust in decision-making, but the high level of fairness statement did not affect user trust.

    “From the results, we can see that, under the feature importance-based explanation condition, the low level of fairness statement significantly decreased the user trust in decision-making, but the high level of fairness statement did not affect user trust.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The introduced fairness did affect the user's perception of fairness, where the lower level of introduced fairness decreased the user's perception of fairness, while the higher level of introduced fairness increased the user's perception of fairness.

    “The results indicated that the introduced fairness did affect the user's perception of fairness, where the lower level of introduced fairness decreased the user's perception of fairness, while the higher level of introduced fairness increased the user's perception of fairness.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Under the low level of introduced fairness, the feature importance-based explanation significantly increased user trust in decision-making, but the example-based explanation did not.

    “Therefore, we can say that, under the low level of introduced fairness, the feature importance-based explanation significantly increased user trust in decision-making, but the example-based explanation did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The effect of explanations on user trust was slightly different in two scenarios of health insurance decision-making and medical treatment decision-making despite explanations increasing user trust in both scenarios.

    “The results indicated that the effect of explanations on user trust was slightly different in two scenarios of health insurance decision-making and medical treatment decisionmaking despite explanations increasing user trust in both scenarios.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Under example-based explanations, the low level of introduced fairness decreased perception of fairness significantly, while the high level of introduced fairness did not.

    “The results indicated that, under example-based explanations, the low level of introduced fairness decreased perception of fairness significantly, while the high level of introduced fairness did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Participants showed slightly different behaviours of perception of fairness in two studied scenarios, where example-based explanations increased perception of fairness in the medical treatment decision-making but not in the health insurance decision-making despite feature importance-based explanations increased perception of fairness in both scenarios.

    “The results revealed that participants showed slightly different behaviours of perception of fairness in two studied scenarios, where example-based explanations increased perception of fairness in the medical treatment decision-making but not in the health insurance decision-making despite feature importance-based explanations increased perception of fairness in both scenarios.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Under feature importance-based explanations, the low level of introduced fairness decreased the perception of fairness significantly, while the high level of introduced fairness did not.

    “Therefore, similar conclusions were obtained for the perception of fairness under feature importance-based explanations as that under example-based explanations: the low level of introduced fairness decreased the perception of fairness significantly, while the high level of introduced fairness did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The effect of introduced fairness on user trust was similar in two scenarios, where the low level of introduced fairness decreased user trust while the high level of introduced fairness did not.

    “The results showed that the effect of introduced fairness on user trust was similar in two scenarios, where the low level of introduced fairness decreased user trust while the high level of introduced fairness did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Under the low level of introduced fairness, feature importance-based explanations significantly increased participants' perception of fairness, while example-based explanations did not.

    “The results showed that, under the low level of introduced fairness, feature importancebased explanations significantly increased participants' perception of fairness, while examplebased explanations did not.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Participants showed similar behaviours of perception of fairness in the two studied scenarios, where the low level of introduced fairness decreased perceived fairness while the high level of introduced fairness did not.

    “The results indicated that participants showed similar behaviours of perception of fairness in two studied scenarios, where the low level of introduced fairness decreased perceived fairness while the high level of introduced fairness did not.”
    • Also stated in: § 5. Discussion · plain
      “While participants showed similar behaviours of perception of fairness in two studied scenarios, where the low level of introduced fairness decreased perceived fairness and the high level of introduced fairness did not.”
    Quote in the paperSubstantial claim
23 of 23

  • The paper relies on this without support.

    Little work exists on whether different explanation types and introduced fairness together affect people's perception of fairness.

    “However, little work is found in understanding whether different explanation types and introduced fairness together affect people's perception of fairness.”

  • The paper relies on this without support.

    Little work exists on how AI explanation and introduced fairness concurrently affect user trust and perception of fairness.

    “Despite the proliferation of investigations of the effects of AI explanation on trust and perception of fairness, or effects of introduced fairness on trust and perception of fairness, it is crucial to understand how AI explanation and introduced fairness concurrently affect user trust and perception of fairness since AI explanation and fairness are common partners in AI-informed decision-making.”

  • The paper relies on this without support.

    AI models are imperfect abstractions of reality because of their statistical nature.

    “AI models are also imperfect abstractions of reality because of their statistical nature.”

  • The paper relies on this without support.

    The data used to train machine learning models are often historical records or samples of events and conceal discrimination.

    “The data used to train machine learning models are often historical records or samples of events.They are usually not a precise description of events and conceal discrimination with sparse details which are very difficult to identify.”
4 of 4

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?

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Not confirmedF(2, 147) = 3.855reported p < .023 · recomputed p = .023

    “F(2, 147) = 3.855, p < .023”

  • Not confirmedF(2, 147) = 2.074reported p < .129 · recomputed p = .129

    “F(2, 147) = 2.074, p < .129”

  • Not confirmedF(1, 448) = 2.263reported p < .133 · recomputed p = .133

    “F(1, 448) = 2.263, p < .133”

  • Not confirmedF(2, 222) = 3.863reported p < .022 · recomputed p = .022

    “F(2, 222) = 3.863, p < .022”

  • ConsistentF(2, 447) = 14.3reported p < 0 · recomputed p < .001

    “F(2, 447) = 14.300, p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 447) = 17.325reported p < 0 · recomputed p < .001

    “F(2, 447) = 17.325, p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(4, 441) = 1.233reported p < .296 · recomputed p = .296

    “F(4, 441) = 1.233, p < .296”

  • ConsistentF(2, 147) = 8.735reported p < 0 · recomputed p < .001

    “F(2, 147) = 8.735, p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 147) = 4.676reported p < .011 · recomputed p = .011

    “F(2, 147) = 4.676, p < .011”

  • ConsistentF(2, 147) = 4.149reported p < .018 · recomputed p = .018

    “F(2, 147) = 4.149, p < .018”

  • ConsistentF(2, 147) = 31.435reported p < 0 · recomputed p < .001

    “F(2, 147) = 31.435,p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 147) = 10.508reported p < 0 · recomputed p < .001

    “F(2, 147) = 10.508,p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(4, 441) = 1.005reported p < .405 · recomputed p = .405

    “F(4, 441) = 1.005, p < .405”

  • ConsistentF(2, 147) = 17.964reported p < 0 · recomputed p < .001

    “F(2, 147) = 17.964, p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 147) = 7.892reported p < .001 · recomputed p < .001

    “F(2, 147) = 7.892, p < .001”

  • ConsistentF(2, 147) = 3.199reported p < .044 · recomputed p = .044

    “F(2, 147) = 3.199, p < .044”

  • ConsistentF(1, 448) = 0.593reported p = .442 · recomputed p = .442

    “F(1, 448) = 0.593, p = .442”

  • ConsistentF(2, 444) = 1.287reported p < .277 · recomputed p = .277

    “F(2, 444) = 1.287, p < .277”

  • ConsistentF(2, 222) = 11.226reported p < 0 · recomputed p < .001

    “F(2, 222) = 11.226,p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 222) = 7.508reported p < .001 · recomputed p < .001

    “F(2, 222) = 7.508, p < .001”

  • ConsistentF(2, 444) = 0.109reported p = .896 · recomputed p = .897

    “F(2, 444) = 0.109, p = .896”

  • ConsistentF(2, 222) = 8.446reported p < 0 · recomputed p < .001

    “F(2, 222) = 8.446, p < .000”

    The paper rounds a very small p to zero

  • ConsistentF(2, 222) = 5.968reported p < .003 · recomputed p = .003

    “F(2, 222) = 5.968, p < .003”

  • ConsistentF(2, 444) = 0.813reported p < .444 · recomputed p = .444

    “F(2, 444) = 0.813, p < .444”

  • ConsistentF(2, 222) = 7.19reported p < .001 · recomputed p < .001

    “F(2, 222) = 7.190, p < .001”

  • ConsistentF(2, 444) = 0.68reported p = .507 · recomputed p = .507

    “F(2, 444) = 0.680, p = .507”

  • ConsistentF(2, 222) = 23.078reported p = 0 · recomputed p < .001

    “F(2, 222) = 23.078,p = .000”

  • ConsistentF(2, 222) = 10.671reported p < 0 · recomputed p < .001

    “F(2, 222) = 10.671,p < .000”

    The paper rounds a very small p to zero

28 of 28

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.

  • Discussion reverses the health-insurance trust result

    Minor

    The discussion says example-based explanations did not increase trust in the health-insurance scenario, whereas the results report significantly higher trust than the control condition in that scenario.

  • Same fairness-scenario trust interaction has two different results

    Minor

    Both passages report the interaction between introduced fairness and scenario on trust, but give different F statistics and p-values for the same degrees of freedom.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Zhou, J., and Chen, F. (2018). Human and Machine Learning: Visible, Explainable, Trustworthy and Transparent, Springer.
  2. Komodromos, 2014, Employees’ Perceptions of Trust, Fairness, and the Management of Change in Three Private Universities in Cyprus, J. Hum. Resour. Manag. Labor Stud., 2, 35
  3. Yin, M., Vaughan, J.W., and Wallach, H. (2018, January 14). Does Stated Accuracy Affect Trust in Machine Learning Algorithms?. Proceedings of the ICML2018 Workshop on Human Interpretability in Machine Learning (WHI 2018), Stockholm, Sweden.

Cited by

Sorted by citations

Related

Authors
Angerschmid, A., Zhou, J., Theuermann, K., Chen, F., & Holzinger, A.
Affiliations
  • Medical University of Graz (AT)
  • University of Technology Sydney (AU)
  • Graz University of Technology (AT)
  • University of Alberta (CA)
Corresponding author
Zhou, Jianlong
ORCID
Venue
Machine Learning and Knowledge Extraction
Publisher
MDPI AG
Published
2022-06-16
Volume
4
Issue
2
Pages
556-579
ISSN
2504-4990 (ISSN-L)
DOI
10.3390/make4020026
OpenAlex
W4283076847
Cited by
161
References
58
Citations by year
2026: 28 · 2025: 40 · 2024: 45 · 2023: 41 · 2022: 7
FWCI
10.94
Type
journal article
Language
en
Open access
Licence
Locations
  • Machine Learning and Knowledge Extraction (publishedVersion, cc-by)
  • OPUS - Open Publications of UTS Scholars (University of Technology Sydney) (submittedVersion, other-oa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Austrian Science Fund (FWF) (P-32554)
  • Australian UTS STEM-HASS Strategic Research Fund 2021 (P-32554)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Psychology › Social Psychology › Human-Automation Interaction and Safety
Keywords
AI-assisted decision-making, AI explanations, fairness, trust in AI, perceived fairness, transparency, treatment decision-making, responsible AI use

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

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