The DOI matches its Crossref record; 87 of the 92 references it cites were found in a registry or an index, and 5 are not checked yet; none of the 27 sources its claims cite is checked for support yet.
'It's Reducing a Human Being to a Percentage': Perceptions of Justice in Algorithmic Decisions
Open accessPublished version of a preprint
Abstract
Data-driven decision-making consequential to individuals raises important questions of accountability and justice. Indeed, European law provides individuals limited rights to 'meaningful information about the logic' behind significant, autonomous decisions such as loan approvals, insurance quotes, and CV filtering. We undertake three experimental studies examining people's perceptions of justice in algorithmic decision-making under different scenarios and explanation styles. Dimensions of justice previously observed in response to human decision-making appear similarly engaged in response to algorithmic decisions. Qualitative analysis identified several concerns and heuristics involved in justice perceptions including arbitrariness, generalisation, and (in)dignity. Quantitative analysis indicates that explanation styles primarily matter to justice perceptions only when subjects are exposed to multiple different styles---under repeated exposure of one style, scenario effects obscure any explanation effects. Our results suggests there may be no 'best' approach to explaining algorithmic decisions, and that reflection on their automated nature both implicates and mitigates justice dimensions.
Checked 2026-10-06 · 5 references not checked yet
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
Three experimental studies found that people apply justice dimensions to algorithmic decisions much as they do to human decisions, but explanation-style effects were mostly visible only when participants compared multiple styles for the same case. Case-based explanations were consistently associated with lower justice perceptions than sensitivity-based explanations, while repeated exposure to a single explanation style largely obscured explanation effects.
Key findings
- Justice constructs correlated similarly to the human decision-making literature: Spearman’s rho was 0.69 for fair process–deserved outcome, 0.60 for fair process–appropriate factors, 0.51 for appropriate factors–deserved outcome, 0.40 for fair process–understanding, 0.40 for appropriate factors–understanding, and 0.36 for understanding–deserved outcome.
- In the between-subjects study (325 participants, 5 conditions of n = 65), explanation styles generally did not significantly affect justice perceptions, except for fair process in loans, F(3, 258) = 2.71, p = 0.046, and appropriate factors, F(3, 258) = 5.35, p = 0.001.
- In the within-subjects study (65 participants), significant explanation effects appeared across all questions and contexts: fair process in loans, F(3, 260) = 7.52, p < .001, and insurance, F(3, 260) = 4.5, p = .004; appropriate factors in loans, F(3, 260) = 3.312, p = .02, and insurance, F(3, 260) = 6.44, p = .0003; deserved outcome in loans, F(3, 260) = 7.31, p = .0001. Case-based explanations produced lower appropriateness, fair process, and loan deservedness than sensitivity-based styles.
- Qualitative themes included lack of human touch/indignity, interpreting the system’s reasoning, acceptability of statistical inference and generalisation, actionability, unaccounted aspects, and the meaning/relevance of moral concepts; some participants saw the algorithmic nature as mitigating justice judgments.
Methods
- Three studies: Phase 1 was an in-person lab study with 19 UK participants, using semi-structured think-aloud interviews over 15 fictional negative-outcome cases across 5 contexts (loan, promotion, car insurance, airline re-routing, bank account freeze), each with one of 4 explanation styles (input influence, sensitivity, case-based, demographic) or a no-explanation control; data were thematically analysed.
- Phase 2 comprised two online Prolific Academic studies with UK participants over 18: a between-subjects study with 325 participants randomly assigned to 5 conditions (n = 65 each) and 12 cases with one explanation style or control, and a within-subjects study with 65 participants using only loan and insurance cases, presenting four different explanation styles for a single case in random order.
- Measures were agreement with five 5-point Likert justice statements (agreement, understanding, appropriateness of factors, fair process, deserved outcome); analysis used Spearman’s rho correlations and ANOVA with Tukey’s post-hoc paired tests.
Limitations
- Stated: samples were not representative—the lab study drew from an affluent, academically dominated UK city, and the online studies were not gender balanced, with almost double the number of females to males; scenarios were hypothetical and did not directly affect participants, lacking first-person consequences and options such as human review or appeal; explanations were not drawn from real machine learning model outputs; differences in explanation text length and format may have confounded results; further confirmatory work is needed on why explanation styles differ.
- Stated/evident: the airline context was excluded from the online studies because lab participants almost universally judged it very unfair and undeserved, preventing useful conclusions; the lab sample was small (19 participants), and online participants were limited to the UK.
- Evident: the between-subjects design repeated a single explanation style, which may have led to habituation, while the within-subjects design made style differences more salient through direct comparison; neither design may fully capture real-world exposure patterns.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
51 Own result · 21 Not checked yet · 17 Own interpretation · 7 Own method · 6 No basis given
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Qualitative analysis identified several concerns and heuristics involved in justice perceptions including arbitrariness, generalisation, and (in)dignity.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Dimensions of justice previously observed in response to human decision-making appear similarly engaged in response to algorithmic decisions.”
hedgedcorrelationaln = 3- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Through iterative testing we found that the use of dialog boxes focused attention on explanations (see Figure 2).”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“In contrast to this apparent lack of humanity (or perhaps to counteract it), some participants attempted to humanise or anthropomorphise the system [48], one participant for instance describing it as 'rude' ( OR ).”
What the sources say
- Not checked yetcites [48] 48. Clifford Nass, Jonathan Steuer, Ellen Tauber, and Heidi Reeder. 1993. Anthropomorphism, agency, and ethopoeia: computers as social actors. In INTERACT'93 and CHI'93 Conference Companion on Human Factors in Computing Systems . ACM, 111-112. source
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In judging the appropriateness of certain features, or the fairness of a decision, issues of personal responsibility and the possibility of acting differently were frequently raised.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Some participants were happy to endorse negative decisions if they were able to identify a reasonable alternative course of action available to the participant which would resulted in a different outcome, as in sensitivity -based explanations:”
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We found that people generally did not pay attention to explanations independent of scenario descriptions.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“An initial reaction from many participants concerned the indignity or 'weirdness' of algorithmic decision-making compared to human decision-making.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We also found that people tended to interpret these justice construct statements as a proxy for the simpler question of whether or not they agreed with the decision;”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, many other participants expressed strong objections to the use of statistical inferences to make judgements about people in itself:”
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
- It may state two separate claims as one. 25
- Probably not a substantial claim. 47
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Quantitative analysis indicates that explanation styles primarily matter to justice perceptions only when subjects are exposed to multiple different styles-under repeated exposure of one style, scenario effects obscure any explanation effects.”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However in this case, some participants were more comfortable with inductive inferences; AS agreed with the computer's decision to deny a loan applicant with existing debts because they had already 'proven not to be able to pay it back'.”
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The airline context was excluded at this stage, because experiment 1 had revealed that participants almost universally judged the cases to be very unfair and undeserved, thus preventing any useful conclusions to be drawn from this context.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Where an individual was deemed at fault for risky or otherwise undesirable past behaviour, participant AS believed the decision was particularly justified:”
plaincorrelationalSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“One participant argued that a truly random system could also be seen as more 'fair', at least for certain definitions of fairness:”
plainSingle case- Asserted: fails — The authors may not put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The authors may not put it forward as true. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the between-subjects study, explanation styles generally did not significantly affect justice perceptions, with the exception of fair process in loans only (loan F(3, 258) = 2.71, p = 0.046) and appropriate factors (F(3, 258) = 5.35, p = 0.001) showing significant between group differences.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
- Rests on an assumption: Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
- Rests on an assumption: The four explanation styles are sufficiently distinct and were correctly perceived by participants as intended.
- Rests on an assumption: The sample of Prolific Academic participants is similar enough to the lab study participants to allow comparison.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In some cases, what was missing was not just information, but also complexity:”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For some participants, the very idea of an algorithmic system making an important decision on the basis of past data seemed unfair;”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In others, the missing element was a sense of proportionality.”
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Whereas for others, delegation of such responsibility to a computer, and the authority of statistical models, actually removed the relevance of questions of fairness altogether; the system 'just does what it is supposed to', and as long as its inferences are accurate, it could be seen as 'statistically fair' ( PC ).”
hedgeddirectSingle case- Asserted: fails — The authors may not put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The authors may not put it forward as true. The paper’s conclusions may not depend on it. It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The meaning and relevance of moral concepts like 'fairness' and 'desert' in the context of a computer system was occasionally questioned by certain participants.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In some cases, while the individual premises seemed understandable, their relevance to the reasoning behind the overall decision was not:”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Issues raised in this regard were the importance of scientific rigour, including the use of sufficient sample sizes:”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Many criticisms of the decisions came down to things that were missing or unaccounted for.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results (see Table 1) partially confirmed our hypotheses, as the stronger correlations were connected to the construct connections we believed would exist.”
plaincorrelationaln = 325- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
- Rests on an assumption: Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In addition to comparisons between different individuals, participants also commented on the acceptability of judging a single individual's future behaviour on the basis of their past behaviour.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Although participant CS was concerned that behaviour change in response to an algorithm could prove dangerous”
hedgedconditionalSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Tukey's post-hoc paired tests (see Figure 3) showed that casebased explanations result in lower perceptions of appropriateness, fair process perception, and (in the loans case) deservedness, consistently compared to sensitivity based styles and occasionally compared to other styles.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
- Rests on an assumption: Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
- Rests on an assumption: The four explanation styles are sufficiently distinct and were correctly perceived by participants as intended.
- Rests on an assumption: The sample of Prolific Academic participants is similar enough to the lab study participants to allow comparison.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Several responses suggested that explanations ought to help guide future action, something which was particularly lacking in the demographic explanations:”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Others tried to articulate the system's reasoning as a series of rules and consequences:”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“There was a significant effect of explanation styles on perceptions of fair process (loan scenario F(3, 260) = 7.52, p < .001; insurance scenario F(3, 260) = 4.5, p = .004), on the perception of appropriate use of factors (loan F(3, 260) = 3.312, p = .02; insurance F(3, 260) = 6.44, p = .0003), and on the perception of deserved outcome (only in the loans case, F(3, 260) = 7.31, p = .0001).”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
- Rests on an assumption: Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
- Rests on an assumption: The four explanation styles are sufficiently distinct and were correctly perceived by participants as intended.
- Rests on an assumption: The sample of Prolific Academic participants is similar enough to the lab study participants to allow comparison.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the same participant, the specificity of the figures used in some explanations (particularly, input influence and sensitivity ) lead her to conclude that the computer must be using arbitrary thresholds, which in itself seemed 'mean':”
plaindirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Many of the reactions concerned attempts to decipher the logic of the algorithm.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Faced with seemingly inexplicable reasoning processes, some participants concluded that the system must be simply 'making it up' ( OR ) and therefore on the face of it unfair.”
plaindirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Others considered it problematic that with an automated system, 'there's no sense of negotiation' ( DR ), and no opportunity for 'human interaction' ( CS ).”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For some, it was the absence of important information:”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participant PC struggled to evaluate the moral dimensions of computer-driven decisions:”
plainSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A series of comments reflected upon the ways in which the very use of an algorithmic decision-making system could be impersonal and dehumanising for the subject of the decision.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Some emphasised that an individual who seems similar to past individuals may in fact have extenuating circumstances, particularly in response to case-based explanations:”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The think-aloud responses from the lab study resulted in five major themes: the lack of human touch; interpretations of the system's reasoning; the use of statistical inference; the degree of actionability in an explanation; important aspects which were seen to be unaccounted for by the system; and the meaning and relevance of moral concepts.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For DR , here was a sense that the very notion of 'fairness' may not be attributable to a system that is primarily designed to serve a different organisational goal like efficiency:”
hedgedSingle case- Asserted: fails — The authors may not put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The authors may not put it forward as true. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Conversely, where a judgement was made on the basis of circumstances which were outside the individual's control, or the course of action suggested was deemed unrealistic, the outcome was seen as undeserved:”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“One participant described how she might have it both ways, simultaneously entertaining dual standards of justice, one juridical, the other computational;”
hedgedSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Such findings provide a rationale for giving individuals rights to information about significant decisions; requiring decision-makers to explain their decisions should promote informational justice, which should help people assess procedural justice and distributive justice in turn.”
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Work on perceptions of justice reveals much about the role of explanations in ethical assessments of human decisions, but the applicability of these findings to algorithmic decisions and the range of proposed explanation systems proposed for them remains largely unexplored.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Such remarks suggest that if algorithmic decisions are held to norms of social behaviour, interactional justice may be a relevant dimension for evaluation.”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
In the paper
“According to Colquitt and others, justice perceptions can be broken down into several aspects or antecedents [10].”
What the sources say
- Not checked yetcites [10] 10. Jason A Colquitt, Donald E Conlon, Michael J Wesson, Christopher OLH Porter, and K Yee Ng. 2001. Justice at the Millennium: A Meta-Analytic Review of 25 Years of Organizational Justice Research. Journal of Applied Psychology 86, 3 (2001). DOI: http://dx.doi.org/10.1037//0021-9010.86.3.425 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“According to the literature on justice perceptions above, we should expect to find correlations between fair process and deserved outcome , understanding and fair process , appropriateness of factors and deserved outcome , and appropriateness of factors and fair process [10].”
What the sources say
- Not checked yetcites [10] 10. Jason A Colquitt, Donald E Conlon, Michael J Wesson, Christopher OLH Porter, and K Yee Ng. 2001. Justice at the Millennium: A Meta-Analytic Review of 25 Years of Organizational Justice Research. Journal of Applied Psychology 86, 3 (2001). DOI: http://dx.doi.org/10.1037//0021-9010.86.3.425 source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is an effect primarily observed, like most effects in the quantitative part of our study, in the within subject study design, indicating that the act of comparison in a particular scenario is important for these differences to become apparent.”
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
- Rests on an assumption: Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
- Rests on an assumption: The four explanation styles are sufficiently distinct and were correctly perceived by participants as intended.
- Rests on an assumption: The sample of Prolific Academic participants is similar enough to the lab study participants to allow comparison.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Thematic codes were independently developed by three coders, after which a single set was jointly agreed upon. The data were then re-coded using the new set of codes.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
None of the sources cited for this claim have been checked yet.
In the paper
“In this sense, while algorithmic decision-making implicates dimensions of justice, it may also mitigate them, in so far as people resist imputing morality to a computer system [25, 5].”
What the sources say
- Not checked yetcites [25] 25. Batya Friedman and Peter H Kahn. 1992. Human agency and responsible computing: Implications for computer system design. Journal of Systems and Software 17, 1 (1992), 7-14. source
- Not checked yetcites [5] 5. Joanna J Bryson. 2010. Robots should be slaves. Close Engagements with Artificial Companions: Key social, psychological, ethical and design issues (2010), 63-74.
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“to avoid this, an additional statement about agreement (1, agreement ) was also included prior to the others.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
None of the sources cited for this claim have been checked yet.
In the paper
“Much of this research is based on field surveys of individuals with personal experience of being affected by decisions (e.g. [79]), although controlled experiments involving simulated decisionmaking and scenario-based methods have also been used [78].”
What the sources say
- Not checked yetcites [79] 79. Kees Van den Bos, Lynn Van der Velden, and Allan Lind. 2014. On the role of perceived procedural justice in citizens' reactions to government decisions and the handling of conflicts. Utrecht Law Review 10, 4 (2014). DOI: http://dx.doi.org/10.18352/ulr.287
- Not checked yetcites [78] 78. Kees Van den Bos, E Allan Lind, Riël V ermunt, and Henk AM Wilke. 1997. How do I judge my outcome when I do not know the outcome of others? The psychology of the fair process effect. Journal of Personality and Social Psychology 72, 5 (1997), 1034. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Requiring organisations to explain the logic behind their algorithmic decision-making systems (informational justice) enables affected individuals to assess whether the logic of the system is just (procedural justice), which in turn might moderate their assessments of fairness of the decision outcomes (distributive justice).”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This was partly due to the hypothetical nature of the scenarios, and partly due to being too human-specific to apply to a computer-based interaction.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Similarly, the United States' Equal Credit Opportunity Act (ECOA) requires lenders to provide 'statements of reasons' when applications are denied loans [32].”
What the sources say
- Not checked yetcites [32] 32. David C Hsia. 1978. Credit scoring and the equal credit opportunity act. Hastings LJ 30 (1978), 371.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Many of the previously mentioned machine learning model explanation systems have not been subjected to user evaluations”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“We also do not yet know much the extent to which findings on correlations between perceptions of justice from human decision-making contexts hold in the context of algorithmic decisions.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“However the nuances of the different explanation approaches recently developed in interpretable ML have not yet been explored in relation to perceptions of justice.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
In the paper
“These aspects of justice are distinct but correlated [10].”
What the sources say
- Not checked yetcites [10] 10. Jason A Colquitt, Donald E Conlon, Michael J Wesson, Christopher OLH Porter, and K Yee Ng. 2001. Justice at the Millennium: A Meta-Analytic Review of 25 Years of Organizational Justice Research. Journal of Applied Psychology 86, 3 (2001). DOI: http://dx.doi.org/10.1037//0021-9010.86.3.425 source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Statement 2 ( understanding ) aimed to capture informational justice [27]; statements 3 and 4 ( appropriateness of factors, fair process ) pertained to procedural justice; and statement 5 ( deserved outcome ) addressed distributive dimensions [11].”
What the sources say
- Not checked yetcites [27] 27. Jerald Greenberg and R Cropanzano. 1993. The social side of fairness: Interpersonal and informational classes of organizational justice. Justice in the workplace: Approaching fairness in human resource management, Lawrence Erlbaum Associates, Hillsdale, NJ (1993).
- Not checked yetcites [11] 11. Jason A Colquitt and Jessica B Rodell. 2015. Measuring justice and fairness. In Oxford Handbook of Justice in the Workplace , Russell Cropanzano and Maureen L Ambrose (Eds.). Oxford University Press, Oxford, 187-202.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
- Rests on an assumption: The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“One possible explanation is that when people repeatedly see the same explanation style, they become habituated and pay more attention to the features of the cases rather than any specific features of the explanation provided.”
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“the scenarios considered were hypothetical, not affecting the participants directly, and therefore lacked the first-person consequences and significance of a real world decision”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The introduction of algorithmic systems seems likely to shake up previously stable categories and concepts, requiring adjustment and incorporation into one's moral world-view.”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
- Rests on an assumption: The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
- Rests on an assumption: The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The algorithmic nature of these systems results in an array of novel considerations which are not captured by traditional research on perceptions of justice.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“people may realistically face multiple explanations for similar algorithmic decisions when applying for a loan, mortgage or credit”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“creators of ML explanation systems have not typically designed them with the information needs of those individuals facing significant personal consequences of model outputs in mind”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When exposed to multiple explanation styles regarding the same case, the differences may become more salient.”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“differences in the length and format of the text required to impart an explanation between explanation conditions may have been a confounding factor”
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The authors state this as an assumption.
The GDPR automated-decision hurdle ('based solely on automated processing') can be read less restrictively than legal analysis has indicated, so the selected scenarios may fall under GDPR.
“The hurdle that a decision must be 'based solely on automated processing' is here read less restrictively than legal analysis has indicated [81, 20].”
The paper relies on this without support.
Explanations are most likely to be requested in response to a decision with a negative outcome for the individual.
“Because explanations are most likely to be requested in response to a decision which has a negative outcome for the individual, and considerations of procedural justice have been found to be more strongly felt in such cases [4], each scenario ended in a negative outcome for the individual.”
The paper relies on this without support.
The justice constructs developed for human decision-making (informational, procedural, distributive) are applicable to and valid for evaluating algorithmic decision-making contexts.
“The justice constructs used in all three studies were based on those developed in the psychology of justice research, summarised by Colquitt et al [11].”
“The remaining questions pertained to informational, procedural and distributive justice.”
The paper relies on this without support.
Participants' Likert-scale responses to the justice statements accurately measure their perceived justice.
“Participants were asked to rate their agreement with five statements on a 5-point Likert scale”
“Participants were asked to indicate their agreement with the 5 measures described in phase 1 using the same Likert scale.”
The paper relies on this without support.
The fictionalised scenarios used in the lab and online studies are realistic and representative enough to elicit valid responses about algorithmic decision-making.
“The first study consisted of in-person semi-structured interviews focused around a series of fictionalised cases as described above.”
“Cases were presented with a brief description of the individual, the decision outcome, the explanation (except in the control condition) and an illustrative image depicting the context.”
The paper relies on this without support.
The four explanation styles are sufficiently distinct and were correctly perceived by participants as intended.
“Each case was presented with one of the 4 explanation styles described above, or a control condition in which no explanation was provided.”
“Each decision was explained using a different style (in a random order), allowing participants to directly compare multiple styles applied to an individual subject.”
The paper relies on this without support.
The thematic analysis themes identified by the coders accurately represent participants' interpretations and reasoning.
“Thematic analysis was undertaken to identify common themes in participants' interpretation, evaluation and reasoning in response to cases. Thematic codes were independently developed by three coders, after which a single set was jointly agreed upon. The data were then re-coded using the new set of codes.”
The paper relies on this without support.
The sample of Prolific Academic participants is similar enough to the lab study participants to allow comparison.
“Participants were recruited via the study platform Prolific Academic, 4 and filtered to include only individuals over 18 based in the UK, in order to maintain similarity with the participants recruited in phase 1.”
The paper relies on this without support.
There is still much to learn about what people want and need to know about algorithms in order to hold them accountable to justice.
“there is still much to learn about what people want and need to know about algorithms in order to hold them accountable to justice”
The paper relies on this without support.
The explanation styles proposed in the interpretable machine learning literature vary in the information they contain.
“the explanation styles proposed in the interpretable machine learning literature vary in the information they contain”
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“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 [17]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Furthermore, humans”
From Artificial Intelligence and Management: The Automation–Augmentation Paradox (2021) · cites this paper as Binns, Van Kleek, Veale, Lyngs, Zhao, & Shadbolt, 2018
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“This philosophically grounded notion is matched by human perceptions that decisions made by algorithms are less trustworthy and fair than those made by humans and that being evaluated by algorithms is dehumanizing ( Binns et al., 2018 ;Lee, 2018 ).”
From The flaws of policies requiring human oversight of government algorithms (2022) · cites this paper as Binns et al., 2018
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“There is evidence that suggests that decision subjects care about justice-related aspects in algorithmic decisionmaking, as they care in human decision-making [17].”
From Disentangling Fairness Perceptions in Algorithmic Decision-Making: the Effects of Explanations, Human Oversight, and Contestability (2023) · cites this paper as [17]
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.
- ConsistentF(3, 258) = 2.71reported p = .046 · recomputed p = .046
“F(3, 258) = 2.71, p = .046”
- ConsistentF(3, 258) = 5.35reported p = .001 · recomputed p = .001
“F(3, 258) = 5.35, p = .001”
- ConsistentF(3, 260) = 4.5reported p = .004 · recomputed p = .004
“F(3, 260) = 4.5, p = .004”
- ConsistentF(3, 260) = 3.312reported p = .02 · recomputed p = .021
“F(3, 260) = 3.312, p = .02”
- ConsistentF(3, 260) = 6.44reported p = .0003 · recomputed p < .001
“F(3, 260) = 6.44, p = .0003”
- ConsistentF(3, 260) = 7.31reported p = .0001 · recomputed p < .001
“F(3, 260) = 7.31, p = .0001”
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.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Do different explanation styles result in differences in perceived levels of justice? · p. 9
“In the within-subjects study-where multiple explanation styles were presented for the same decision-significant effects were observed across all questions and decision-making contexts.”
§ Do different explanation styles result in differences in perceived levels of justice? · p. 9
“on the perception of deserved outcome (only in the loans case, F(3, 260) = 7.31, p = .0001).”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Quantitative results · p. 8
“we calculated Spearman's rho correlation coefficients (with 0.05 confidence level bootstrapped intervals).”
§ Do different explanation styles result in differences in perceived levels of justice? · p. 9
“Table 1. Spearman's rho correlations for justice constructs with bootstrapped (1000 rep) confidence interval at 0.02 level.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Qualitative Results: Themes and reflections · p. 6
“The think-aloud responses from the lab study resulted in five major themes”
§ Qualitative Results: Themes and reflections · p. 6
“the lack of human touch; interpretations of the system's reasoning; the use of statistical inference; the degree of actionability in an explanation; important aspects which were seen to be unaccounted for by the system; and the meaning and relevance of moral concepts.”
Citations
Cites
"Why Should I Trust You?": Explaining the Predictions of Any Classifier
Ribeiro, M. T., Singh, S., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939778 · Cited by 17,555 · cited 4×
“are applicable to any kind of learning algorithm, such as LIME [62], rather than methods which only work for particular ML models (e.”
BackgroundMethodInfluential“Why Should I Trust You?”: Explaining the Predictions of Any Classifier
Ribeiro, M., Singh, S., & Guestrin, C. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations · doi:10.18653/v1/n16-3020 · Cited by 6,682
The Social Psychology of Procedural Justice
Lind, E. A., & Tyler, T. R. (1988) · Critical Issues in Social Justice · doi:10.1007/978-1-4899-2115-4 · Cited by 6,585 · cited 2×
“A significant body of research in legal and organisational psychology suggests that people do not only care about whether the outcome of a decision benefits them, but also whether it meets standards of justice [71, 40, 11].”
BackgroundJustice at the millennium: A meta-analytic review of 25 years of organizational justice research.
Colquitt, J. A., Conlon, D. E., Wesson, M. J., Porter, C. O. L. H., & Ng, K. Y. (2001) · Journal of Applied Psychology · doi:10.1037/0021-9010.86.3.425 · Cited by 5,942 · cited 5×
“According to the literature on justice perceptions above, we should expect to find correlations between fair process and deserved outcome , understanding and fair process , appropriateness of factors and deserved outcome , and appropriateness of factors and fair process [10].”
BackgroundResultInfluentialLipton, Z. C. (2018) · Queue · doi:10.1145/3236386.3241340 · Cited by 5,427 · cited 3×
“2Although we acknowledge this is an important challenge if explanations are to be more than just comforting stories, as cautioned by Lipton [41].”
BackgroundLikert scales, levels of measurement, and the “laws” of statistics
Norman, G. (2010) · Advances in Health Sciences Education · doi:10.1007/s10459-010-9222-y · Cited by 4,850 · cited 1×
“6Some readers might have reservations over parametric tests being applied to ordinal data: we point to recent literature that has emphasised [53] that Likert data are robust to parametric assumptions, and that ‘parametric methods can be utilized without concern for “getting the wrong answer”’.”
BackgroundComparing Individual Means in the Analysis of Variance
Tukey, J. W. (1949) · Biometrics · doi:10.2307/3001913 · Cited by 4,302 · cited 1×
“We used ANOVA tests, followed by Tukey’s post-hoc paired tests [75], to analyse the relative effects of different explanation styles within different study designs.”
MethodBias in random forest variable importance measures: Illustrations, sources and a solution
Strobl, C., Boulesteix, A., Zeileis, A., & Hothorn, T. (2007) · BMC Bioinformatics · doi:10.1186/1471-2105-8-25 · Cited by 3,970
How the machine “thinks”: Understanding opacity in machine learning algorithms
Burrell, J. (2016) · Big Data & Society · doi:10.1177/2053951715622512 · Cited by 3,196
Towards A Rigorous Science of Interpretable Machine Learning
Doshi-Velez, F., & Kim, B. (2017) · arXiv · doi:10.48550/arxiv.1702.08608 · Cited by 3,121 · cited 2×
“Furthermore, as research into practically achievable explanations for machine learning systems accelerates, researchers are turning to consider which explanation styles are more desirable, rather than simply possible to implement [41, 15].”
BackgroundWeapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy
Doyle, T. (2017) · The Information Society · doi:10.1080/01972243.2017.1354593 · Cited by 3,111 · cited 1×
“3 After consulting relevant literature on how ML systems are being applied in each context (e.g. [43, 38, 43, 55, 72]), we settled on the following contexts: 1.”
BackgroundNass, C., Steuer, J., & Tauber, E. R. (1994) · Proceedings of the SIGCHI Conference on Human Factors in Computing Systems · doi:10.1145/191666.191703 · Cited by 2,148 · cited 1×
“On the other hand, people frequently interact with computers as if they were social actors [49]; why stop when justice is involved? One participant described how she might have it both ways, simultaneously entertaining dual standards of justice, one juridical, the other computational;”
Certifying and Removing Disparate Impact
Feldman, M., Friedler, S. A., Moeller, J., Scheidegger, C., & Venkatasubramanian, S. (2015) · Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2783258.2783311 · Cited by 1,827 · cited 1×
“Pr epr in t correct discrimination in data mining exist, these are aimed at data scientists implementing systems rather providing information to the individual decision-subjects affected by them [3, 59, 29, 23, 28].”
BackgroundExplaining collaborative filtering recommendations
Herlocker, J. L., Konstan, J. A., & Riedl, J. (2000) · Proceedings of the 2000 ACM conference on Computer supported cooperative work · doi:10.1145/358916.358995 · Cited by 1,731 · cited 2×
“In some cases, explanation increases trust and reliance [33, 85, 1, 30, 60], but in others an explanation may have the opposite effect if the level of detail it contains is deemed insufficient [7].”
BackgroundBig data: a revolution that will transform how we live, work, and think
Armstrong, K. (2014) · Information Communication & Society · doi:10.1080/1369118x.2014.923482 · Cited by 1,645
Some observations on mental models
Norman, D. A. (1983) · Medical Entomology and Zoology · Cited by 1,549
The price is unfair! A conceptual framework of price fairness perceptions
Xia, L., Monroe, K. B., & Cox, J. L. (2004) · Journal of Marketing · doi:10.1509/jmkg.68.4.1.42733 · Cited by 1,474 · cited 1×
“Research on justice judgements in non-automated contexts has found that individuals pay more attention to other people’s outcomes than they do to information about the procedure behind the decision that affects them [78, 84].”
BackgroundBrockner, J., & Wiesenfeld, B. M. (1996) · Psychological Bulletin · doi:10.1037/0033-2909.120.2.189 · Cited by 1,368 · cited 1×
“Because explanations are most likely to be requested in response to a decision which has a negative outcome for the individual, and considerations of procedural justice have been found to be more strongly felt in such cases [4], each scenario ended in a negative outcome for the individual.”
BackgroundExplaining nonlinear classification decisions with deep taylor decomposition
Montavon, G., Lapuschkin, S., Binder, A., Samek, W., & Müller, K.-R. (2017) · Pattern Recognition · doi:10.1016/j.patcog.2016.11.008 · Cited by 1,357 · cited 2×
“1Layer-wise Relevance Propagation/Deep Taylor Expansion [45], decomposes the innards of deep learning systems to provide similar results to LIME (without directionality), and is also relevant here.”
BackgroundResultWachter, S., Mittelstadt, B., & Floridi, L. (2017) · International Data Privacy Law · doi:10.1093/idpl/ipx005 · Cited by 1,308
Benchmarking state-of-the-art classification algorithms for credit scoring: an update of research
Lessmann, S., Baesens, B., Seow, H.-V., & Thomas, L. C. (2015) · European Journal of Operational Research · doi:10.1016/j.ejor.2015.05.030 · Cited by 1,284 · cited 1×
“[43, 38, 43, 55, 72]), we settled on the following contexts:”
The role of trust in automation reliance
Dzindolet, M. T., Peterson, S. A., Pomranky, R. A., Pierce, L. G., & Beck, H. P. (2003) · International Journal of Human-Computer Studies · doi:10.1016/s1071-5819(03)00038-7 · Cited by 1,276 · cited 1×
“Various research suggests that users—again, primarily experts— may in certain cases systematically over-rely on the outputs of intelligent systems [26, 19], while in others they may systematically distrust systems, despite reassurances as to their accuracy [57, 24].”
BackgroundEvaluating the Visualization of What a Deep Neural Network Has Learned
Samek, W., Binder, A., Montavon, G., Lapuschkin, S., & Muller, K.-R. (2017) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2016.2599820 · Cited by 1,164
Automation bias: a systematic review of frequency, effect mediators, and mitigators
Goddard, K., Roudsari, A., & Wyatt, J. C. (2012) · Journal of the American Medical Informatics Association · doi:10.1136/amiajnl-2011-000089 · Cited by 1,162 · cited 1×
“Various research suggests that users—again, primarily experts— may in certain cases systematically over-rely on the outputs of intelligent systems [26, 19], while in others they may systematically distrust systems, despite reassurances as to their accuracy [57, 24].”
BackgroundThe social side of fairness: Interpersonal and informational classes of organizational justice.
Greenberg, J. (1993) · Cited by 966
MUIR, B. M. (1994) · Ergonomics · doi:10.1080/00140139408964957 · Cited by 873 · cited 1×
“As users form mental models [52] of machine learning systems over time, such long-term exposure may lead to greater understanding [76] and greater trust and reliance [47].”
BackgroundDatta, A., Sen, S., & Zick, Y. (2016) · 2016 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp.2016.42 · Cited by 728 · cited 2×
“Input Influence : Presents a list of input variables alongside a quantitative measure of their ‘influence’, positive or negative, on a decision (e.g. [13, 86]) 2.”
BackgroundMethodThe epistemology of a rule-based expert system —a framework for explanation
Clancey, W. J. (1983) · Artificial Intelligence · doi:10.1016/0004-3702(83)90008-5 · Cited by 712 · cited 2×
“Early knowledge representation and reasoning systems typically aimed to meet this requirement by producing explanations for their outputs, for instance by presenting derivations from rules in rule-based architecture [69, 8], or graphical representations of nodes and edges in the case of Bayesian networks [37].”
BackgroundDiscrimination-aware data mining
Pedreshi, D., Ruggieri, S., & Turini, F. (2008) · Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/1401890.1401959 · Cited by 676 · cited 1×
“Pr epr in t correct discrimination in data mining exist, these are aimed at data scientists implementing systems rather providing information to the individual decision-subjects affected by them [3, 59, 29, 23, 28].”
BackgroundHow Much Information?: Effects of Transparency on Trust in an Algorithmic Interface
Kizilcec, R. F. (2016) · Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems · doi:10.1145/2858036.2858402 · Cited by 615 · cited 1×
“A notable exception is Kizilcec, who studied the effect of levels of transparency on perceptions of procedural justice [35].”
BackgroundNass, C., Steuer, J., & Tauber, E. R. (1994) · Conference companion on Human factors in computing systems - CHI '94 · doi:10.1145/259963.260288 · Cited by 579
Why and why not explanations improve the intelligibility of context-aware intelligent systems
Lim, B. Y., Dey, A. K., & Avrahami, D. (2009) · Proceedings of the SIGCHI Conference on Human Factors in Computing Systems · doi:10.1145/1518701.1519023 · Cited by 552 · cited 2×
“Inspired by different theoretical models of argumentation, various explanation types have been explored and tested, with the style of explanation provided affecting the extent to which end-users understand the decisions of expert systems [46, 39].”
BackgroundEdwards, L., & Veale, M. (2017) · doi:10.31228/osf.io/97upg · Cited by 524
Self-citationDirect manipulation vs. interface agents
Shneiderman, B., & Maes, P. (1997) · Interactions · doi:10.1145/267505.267514 · Cited by 518 · cited 1×
“As a general principle, systems should be able account for their own operation in ways that help users understand how their tasks are being accomplished [16, 66].”
BackgroundFildes, R. A., Goodwin, P., Lawrence, M. J., & Nikolopoulos, K. I. (2009) · International Journal of Forecasting · doi:10.1016/j.ijforecast.2008.11.010 · Cited by 447 · cited 1×
“Various research suggests that users—again, primarily experts— may in certain cases systematically over-rely on the outputs of intelligent systems [26, 19], while in others they may systematically distrust systems, despite reassurances as to their accuracy [57, 24].”
BackgroundThe Role of Explanations on Trust and Reliance in Clinical Decision Support Systems
Bussone, A., Stumpf, S., & O'Sullivan, D. (2015) · 2015 International Conference on Healthcare Informatics · doi:10.1109/ichi.2015.26 · Cited by 441 · cited 1×
“In some cases, explanation increases trust and reliance [33, 85, 1, 30, 60], but in others an explanation may have the opposite effect if the level of detail it contains is deemed insufficient [7].”
BackgroundTickle, A. B., Andrews, R., Golea, M., & Diederich, J. (1998) · IEEE Transactions on Neural Networks · doi:10.1109/72.728352 · Cited by 437 · cited 2×
“The increasing popularity of machine learning models as decision making tools presents novel challenges for explanation, as it may not be as straightforward to extract the embedded ‘knowledge’ behind their outputs [73].”
BackgroundThe Oxford Handbook of Justice in the Workplace
(2015) · Oxford University Press eBooks · doi:10.1093/oxfordhb/9780199981410.001.0001 · Cited by 419
van den Bos, K., Lind, E. A., Vermunt, R., & Wilke, H. A. M. (1997) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.72.5.1034 · Cited by 392
UX Design Innovation: Challenges for Working with Machine Learning as a Design Material
Dove, G., Halskov, K., Forlizzi, J., & Zimmerman, J. (2017) · Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3025453.3025739 · Cited by 391 · cited 1×
“This is part of a wider need to consider how to re-engineer better user experiences for multiple stakeholders interacting with machine learning systems in general [17].”
BackgroundIntrigue: Personalized recommendation of tourist attractions for desktop and hand held devices
Ardissono, L., Goy, A., Petrone, G., Segnan, M., & Torasso, P. (2003) · Applied Artificial Intelligence · doi:10.1080/713827254 · Cited by 384 · cited 2×
“rom the model’s training data which most closely resemble the output in question [12,54], while ‘demographic’ explanations provide statistics on the outcome classes for relevant demographic variables [2]. While these different explanation stylessuggest promising ways to explain the outputs of any classifier, explanati…”
BackgroundFirst I "like" it, then I hide it: Folk Theories of Social Feeds
Eslami, M., Karahalios, K., Sandvig, C., Vaccaro, K., Rickman, A., Hamilton, K., & Kirlik, A. (2016) · Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems · doi:10.1145/2858036.2858494 · Cited by 375 · cited 1×
“But in the absence of transparency from platform providers, end-users may develop folk theories of algorithms, leading to a diverse range of strategies and countermeasures with varying degrees of effectiveness [22, 14].”
BackgroundExplaining Recommendations: Design and Evaluation
Tintarev, N., & Masthoff, J. (2015) · Springer eBooks · doi:10.1007/978-1-4899-7637-6_10 · Cited by 366 · cited 1×
“Tintarev and Mastho identify seven purposes for recommender system explanations, namely: transparency, scrutability, trust, effectiveness, persuasiveness, efficiency and satisfaction [74].”
BackgroundThe relative influence of advice from human experts and statistical methods on forecast adjustments
Önkal, D., Goodwin, P., Thomson, M. E., Gonul, M. S., & Pollock, A. C. (2009) · Journal of Behavioral Decision Making · doi:10.1002/bdm.637 · Cited by 331 · cited 1×
“Various research suggests that users—again, primarily experts— may in certain cases systematically over-rely on the outputs of intelligent systems [26, 19], while in others they may systematically distrust systems, despite reassurances as to their accuracy [57, 24].”
BackgroundAn analysis of physician attitudes regarding computer-based clinical consultation systems
Teach, R. L., & Shortliffe, E. H. (1981) · Computers and Biomedical Research · doi:10.1016/0010-4809(81)90012-4 · Cited by 328 · cited 1×
“Early knowledge representation and reasoning systems typically aimed to meet this requirement by producing explanations for their outputs, for instance by presenting derivations from rules in rule-based architecture [69, 8], or graphical representations of nodes and edges in the case of Bayesian networks [37].”
BackgroundProfiling the European Citizen
Hildebrandt, M., & Gutwirth, S. (2008) · Springer eBooks · doi:10.1007/978-1-4020-6914-7 · Cited by 322
Getting access to what goes on in people's heads?
Nielsen, J., Clemmensen, T., & Yssing, C. (2002) · Nordic Conference on Human-Computer Interaction · doi:10.1145/572020.572033 · Cited by 318 · cited 1×
“The cases were used to prompt reflection on various aspects of the decision, and any explanatory information provided, using a concurrent think-aloud process [51].”
MethodDeVito, M. A., Gergle, D., & Birnholtz, J. (2017) · Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3025453.3025659 · Cited by 311 · cited 1×
“But in the absence of transparency from platform providers, end-users may develop folk theories of algorithms, leading to a diverse range of strategies and countermeasures with varying degrees of effectiveness [22, 14].”
BackgroundCredit Scoring and Its Applications
Thomas, L. C., Crook, J. N., & Edelman, D. B. (2017) · Cited by 308
Vig, J., Sen, S., & Riedl, J. (2009) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/1502650.1502661 · Cited by 304
Trust building with explanation interfaces
Pu, P., & Chen, L. (2006) · Proceedings of the 11th international conference on Intelligent user interfaces · doi:10.1145/1111449.1111475 · Cited by 300 · cited 1×
“In some cases, explanation increases trust and reliance [33, 85, 1, 30, 60], but in others an explanation may have the opposite effect if the level of detail it contains is deemed insufficient [7].”
BackgroundThe Impact of Explanation Facilities on User Acceptance of Expert Systems Advice
Ye, L. R., & Johnson, P. E. (1995) · MIS Quarterly · doi:10.2307/249686 · Cited by 290 · cited 1×
“In some cases, explanation increases trust and reliance [33, 85, 1, 30, 60], but in others an explanation may have the opposite effect if the level of detail it contains is deemed insufficient [7].”
BackgroundSlave to the Algorithm? Why a Right to Explanationn is Probably Not the Remedy You are Looking for
Edwards, L., & Veale, M. (2017) · SSRN Electronic Journal · doi:10.2139/ssrn.2972855 · Cited by 244 · cited 5×
“ncluding both technical work on interpretable machine learning models, and legal discussions on the requirements of existing and forthcoming regulations on transparency of algorithmic decisions (e.g. [81,65,20]. Our aim was to find explanation styles which could plausibly meet or exceed the regulatory requirements reg…”
BackgroundMethodInfluentialSelf-citationA review of explanation methods for Bayesian networks
Lacave, C., & Díez, F. J. (2002) · The Knowledge Engineering Review · doi:10.1017/s026988890200019x · Cited by 240 · cited 1×
“Early knowledge representation and reasoning systems typically aimed to meet this requirement by producing explanations for their outputs, for instance by presenting derivations from rules in rule-based architecture [69, 8], or graphical representations of nodes and edges in the case of Bayesian networks [37].”
BackgroundInterpretable classification models for recidivism prediction
Zeng, J., Ustun, B., & Rudin, C. (2017) · Journal of the Royal Statistical Society Series A: Statistics in Society · doi:10.1111/rssa.12227 · Cited by 220
Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy
Verma, S. (2019) · Vikalpa: The Journal for Decision Makers · doi:10.1177/0256090919853933 · Cited by 219
Enhanced Maintenance and Explanation of Expert Systems Through Explicit Models of Their Development
Neches, R. T., Swartout, W., & Moore, J. D. (1985) · IEEE Transactions on Software Engineering · doi:10.1109/tse.1985.231882 · Cited by 165 · cited 2×
“Explanation facilities could be further augmented by explicitly modeling the knowledge engineering and design process and revealing information about them to end-users [50].”
BackgroundProcedural Justice as Fairness
Thibaut, J., Walker, L., LaTour, S., & Houlden, P. (1974) · Stanford Law Review · doi:10.2307/1227990 · Cited by 164
Credit Scoring and Its Applications, Second Edition
Thomas, L. C., Crook, J. N., & Edelman, D. (2017) · Society for Industrial and Applied Mathematics eBooks · doi:10.1137/1.9781611974560 · Cited by 164
Anthropomorphism, agency, and ethopoeia: computers as social actors
Nass, C., Steuer, J., Tauber, E., & Reeder, H. (1993) · INTERACT '93 and CHI '93 conference companion on Human factors in computing systems - CHI '93 · doi:10.1145/259964.260137 · Cited by 162 · cited 1×
“In contrast to this apparent lack of humanity (or perhaps to counteract it), some participants attempted to humanise or anthropomorphise the system [48], one participant for instance describing it as ‘rude’ (OR).”
ResultMeasuring Justice and Fairness
Cropanzano, R., Ambrose, M. L., Colquitt, J. A., & Rodell, J. B. (2015) · Oxford University Press eBooks · doi:10.1093/oxfordhb/9780199981410.013.8 · Cited by 160
Williams, J. J., Kim, J., Rafferty, A. N., Maldonado, S., Gajos, K. Z., Lasecki, W. S., & Heffernan, N. T. (2016) · ACM Conference on Learning @ Scale (L@S) · doi:10.1145/2876034.2876042 · Cited by 147 · cited 1×
“Contexts studied include aiding medical experts in diagnosis, increasing test performance of users of personalised learning environments, or enabling network security analysts to correctly identify incidents (e.g. [34, 83, 21]).”
BackgroundTullio, J., Dey, A. K., Chalecki, J., & Fogarty, J. A. (2007) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/1240624.1240630 · Cited by 137 · cited 1×
“As users form mental models [52] of machine learning systems over time, such long-term exposure may lead to greater understanding [76] and greater trust and reliance [47].”
BackgroundThe Case for Process Fairness in Learning: Feature Selection for Fair Decision Making
Grgić-Hlača, N., Zafar, M. B., Gummadi, K. P., & Weller, A. (2016) · MPG.PuRe (Max Planck Society) · Cited by 137
Some Observations on Mental Models
Norman, D. A. (2014) · Psychology Press eBooks · doi:10.4324/9781315802725-5 · Cited by 137 · cited 1×
“As users form mental models [52] of machine learning systems over time, such long-term exposure may lead to greater understanding [76] and greater trust and reliance [47].”
BackgroundHuman agency and responsible computing: Implications for computer system design
Friedman, B., & Kahn, P. H. (1992) · Journal of Systems and Software · doi:10.1016/0164-1212(92)90075-u · Cited by 123 · cited 1×
“In this sense, while algorithmic decision-making implicates dimensions of justice, it may also mitigate them, in so far as people resist imputing morality to a computer system [25, 5].”
BackgroundAn Evaluation of the Usefulness of Case-Based Explanation
Cunningham, P., Doyle, D., & Loughrey, J. (2007) · Springer eBooks · doi:10.1007/3-540-45006-8_12 · Cited by 118
Towards improving trust in context-aware systems by displaying system confidence
Antifakos, S., Kern, N., Schiele, B., & Schwaninger, A. (2005) · International Conference on Human-Computer Interaction with Mobile Devices and Services (MobileHCI) · doi:10.1145/1085777.1085780 · Cited by 103
(2017) · arXiv (Cornell University) · Cited by 86
A Case-Based Explanation System for Black-Box Systems
Nugent, C., & Cunningham, P. (2005) · Artificial Intelligence Review · doi:10.1007/s10462-005-4609-5 · Cited by 84 · cited 1×
“1 Case-based explanations provide examples from the model’s training data which most closely resemble the output in question [12, 54], while ‘demographic’ explanations provide statistics on the outcome classes for relevant demographic variables [2].”
BackgroundBureaucratic Justice: Managing Social Security Disability Claims
Hahn, H., & Mashaw, J. L. (1984) · Public Administration Review · doi:10.2307/975875 · Cited by 81
User Preferences for Hybrid Explanations
Kouki, P., Schaffer, J., Pujara, J., O’Donovan, J., & Getoor, L. C. (2017) · ACM Conference on Recommender Systems (RecSys) · doi:10.1145/3109859.3109915 · Cited by 73 · cited 1×
“While early recommender system explanation approaches provided a uniform explanation style for single-source collaborative filtering [30], more recent work explores how to derive explanations for recommender systems based on ‘hybrid’ multiple sources [36] and matrix factorisation [61].”
BackgroundExplanation facilities and interactive systems
Johnson, H., & Johnson, P. (1993) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/169891.169951 · Cited by 71 · cited 1×
“In some cases, explanation increases trust and reliance [33, 85, 1, 30, 60], but in others an explanation may have the opposite effect if the level of detail it contains is deemed insufficient [7].”
BackgroundA Reactive Approach to Explanation: Taking the User’s Feedback into Account
Moore, J. D., & Swartout, W. (1991) · Kluwer international series in engineering and computer science · doi:10.1007/978-1-4757-5945-7_1 · Cited by 67 · cited 1×
“Inspired by different theoretical models of argumentation, various explanation types have been explored and tested, with the style of explanation provided affecting the extent to which end-users understand the decisions of expert systems [46, 39].”
Backgroundvan den Bos, K., van der Velden, L., & Lind, E. A. (2014) · Utrecht Law Review · doi:10.18352/ulr.287 · Cited by 63 · cited 2×
“[79]), although controlled experiments involving simulated decisionmaking and scenario-based methods have also been used [78].”
BackgroundA Review of Explanation and Explanation in Case-Based Reasoning
Doyle, D., Tsymbal, A., & Cunningham, P. (2003) · Trinity's Access to Research Output (TARA) (Trinity College Dublin) · Cited by 61
Credit Scoring and the Equal Credit Opportunity Act
Hsia, D. C. (1978) · bepress Legal Repository · Cited by 56
Moral coherence processes: constructing culpability and consequences
Clark, C. J., Chen, E. E., & Ditto, P. H. (2015) · Current Opinion in Psychology · doi:10.1016/j.copsyc.2015.07.016 · Cited by 50 · cited 1×
“Such attempts to delineate moral culpability between systems and people, or between different standards of fairness or evidence, may reflect a more general psychological need to see the world as morally coherent [9].”
BackgroundHuman Agency and Responsible Computing: Implications for Computer System Design
Friedman, B., & Kahn, P. H. (2017) · Routledge eBooks · doi:10.4324/9781315259697-26 · Cited by 46
Berendt, B., & Preibusch, S. (2014) · Artificial Intelligence and Law · doi:10.1007/s10506-013-9152-0 · Cited by 44 · cited 2×
“Pr epr in t correct discrimination in data mining exist, these are aimed at data scientists implementing systems rather providing information to the individual decision-subjects affected by them [3, 59, 29, 23, 28].”
BackgroundTaking advice from intelligent systems: the double-edged sword of explanations
Ehrlich, K., Kirk, S. E., Patterson, J., Rasmussen, J. C., Ross, S. I., & Gruen, D. (2011) · International Conference on Intelligent User Interfaces · doi:10.1145/1943403.1943424 · Cited by 42 · cited 1×
“Contexts studied include aiding medical experts in diagnosis, increasing test performance of users of personalised learning environments, or enabling network security analysts to correctly identify incidents (e.g. [34, 83, 21]).”
BackgroundTintarev, N. (2007) · Lecture notes in computer science · doi:10.1007/978-3-540-73078-1_67 · Cited by 38
Evaluating the visualization of what a Deep Neural Network has learnedPreprint
Samek, W., Binder, A., Montavon, G., Bach, S., & Müller, K. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1509.06321 · Cited by 35 · cited 1×
“Sensitivity : For each input variable used in a decision, sensitivity analysis shows how much the value of that variable would have to differ in order to change the output class (not to be confused with the notion of sensitivity used in ML evaluation) (e.g. [63, 65]).”
MethodVehicle Telematics at an Italian Insurer: New Auto Insurance Products and a New Industry Ecosystem
Vaia, G., Carmel, E., DeLone, W. H., Trautsch, H., & Menichetti, F. (2012) · Journal of the Association for Information Systems · Cited by 32
Direct and Indirect Discrimination Prevention Methods
Hajian, S., & Domingo-Ferrer, J. (2013) · Studies in Applied Philosophy, Epistemology and Rational Ethics · doi:10.1007/978-3-642-30487-3_13 · Cited by 27 · cited 1×
“Pr epr in t correct discrimination in data mining exist, these are aimed at data scientists implementing systems rather providing information to the individual decision-subjects affected by them [3, 59, 29, 23, 28].”
BackgroundDo News Consumers Want Explanations for Personalized News Rankings?
ter Hoeve, M., Heruer, M., Odijk, D., Schuth, A., Spitters, M., & de Rijke, M. (2018) · doi:10.18122/b24d7n · Cited by 20 · cited 1×
“However, a reported need for explanation may not always correspond to differences in behaviour or performance; in a study of news recommender systems, end-users expressed a desire for explanations, but the number of news items they opened did not change when provided with reasons for their recommendations [70].”
BackgroundSchwartz, P. M. (1991) · bepress Legal Repository · Cited by 16
FATREC Workshop on Responsible Recommendation Proceedings
Ekstrand, M. D., & Sharma, A. K. (2017) · Scholar Works (Boise State University) · Cited by 11
Exploring Explanations for Matrix Factorization Recommender Systems
Rastegarpanah, B., Crovella, M. E., & Gummadi, K. P. (2018) · doi:10.18122/b2r717 · Cited by 11 · cited 1×
“While early recommender system explanation approaches provided a uniform explanation style for single-source collaborative filtering [30], more recent work explores how to derive explanations for recommender systems based on ‘hybrid’ multiple sources [36] and matrix factorisation [61].”
BackgroundKim, B., Chacha, C. M., & Shah, J. (2013) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v27i1.8548 · Cited by 8 · cited 1×
“Contexts studied include aiding medical experts in diagnosis, increasing test performance of users of personalised learning environments, or enabling network security analysts to correctly identify incidents (e.g. [34, 83, 21]).”
BackgroundInferring team task plans from human meetings
KimBeen, ChachaCaleb, M., & ShahJulie, A. (2015) · doi:10.5555/2831407.2831415 · Cited by 3
Hendrickx, F. (2000) · Cited by 2
Research works printed but not linked: 8
- Joanna J Bryson. 2010. Robots should be slaves. Close Engagements with Artificial Companions: Key social, psychological, ethical and design issues (2010), 63--74.
- 10.5555/270318.270324 · DOI
- 10.5555/1386322 · DOI
- Zachary C Lipton. 2016. The Mythos of Model Interpretability. In 2016 ICML Workshop on Human Interpretability in Machine Learning (WHI 2016). New York. https://arxiv.org/abs/1606.03490 · arXiv
- Gregory Mitchell. 2004. Libertarian Paternalism is an Cxymoron. Nw. UL Rev. 99 (2004), 1245.
- 10.5555/3175762 · DOI
- Cathy O'Neill. 2016. How algorithms rule our working lives. The Guardian (2016).
- Frank Pasquale. 2015. Digital Star Chamber. Aeon (2015). https://aeon.co/essays/ judge-jury-and-executioner-the-unaccountable-algorithm
Cited by
Sorted by citations
Artificial Intelligence and Management: The Automation–Augmentation Paradox
Raisch, S., & Krakowski, S. (2021) · Academy of Management Review · doi:10.5465/amr.2018.0072 · Cited by 2,755
Ali, S., Abuhmed, T., El-Sappagh, S., Muhammad, K., Alonso-Moral, J. M., Confalonieri, R., Guidotti, R., Del Ser, J., Díaz-Rodríguez, N., & Herrera, F. (2023) · Information Fusion · doi:10.1016/j.inffus.2023.101805 · Cited by 1,762
Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F., Breazeal, C., Crandall, J. W., Christakis, N. A., Couzin, I. D., Jackson, M. O., et al. (2019) · Nature · doi:10.1038/s41586-019-1138-y · Cited by 1,364
Morley, J., Floridi, L., Kinsey, L., & Elhalal, A. (2020) · Science and Engineering Ethics · doi:10.1007/s11948-019-00165-5 · Cited by 882
“…a system is explicable (explainable and interpretable) it is inherently more transparent and therefore more accountable in terms of its decision-making properties and the extent to which they include human oversight and are fair, robust and justifiable (Binns et al. 2018; Cath 2018; Lipton 2016).”
BackgroundImproving Fairness in Machine Learning Systems: What Do Industry Practitioners Need?
Holstein, K., Wortman Vaughan, J., Daumé, H., Dudik, M., & Wallach, H. (2019) · Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3290605.3300830 · Cited by 777
“For example, HCI researchers have empirically studied users’ expectations and perceptions related to fairness in algorithmic systems, finding that these do not always align with existing statistical definitions [16, 71, 72, 108].”
Background
Metadata
- Authors
- Binns, R., Van Kleek, M., Veale, M., Lyngs, U., Zhao, J., & Shadbolt, N.
- Affiliations
- University of Oxford (GB)
- University of Oxford, Oxford, Oxfordshire , United Kingdom
- University College London (GB)
- Venue
- Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems
- Publisher
- ACM
- Published
- 2018-04-21
- Pages
- 1-14
- OpenAlex
- W4296978576
- Semantic Scholar
- aba609e6c5c9b58f9dac272b75983f75a7f190e5
- Cited by
- 691
- References
- 92
- Citations by year
- 2026: 139 · 2025: 204 · 2024: 87 · 2023: 60 · 2022: 76 · 2021: 53 · 2020: 35 · 2019: 31 · 2018: 6
- FWCI
- 56.17
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- CHI Conference on Human Factors in Computing Systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Funders
- UK Engineering and Physical Sciences Research Council (EP/J017728/2)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Social Sciences › Law › Artificial Intelligence in Law
- Keywords
- algorithmic decision-making, perceived justice, algorithmic transparency, explanation styles, accountability, arbitrariness, indignity, generalization, automated decision-making, EU data protection law, lending, insurance quotes
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.
A language model drew the claims from the paper on 2026-10-06, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-06 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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
