The DOI matches its Crossref record; 6 of the 6 references it cites were found in a registry or an index; none of the 85 sources its claims cite is checked for support yet.
Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI
Open accessAbstract
This paper examines the legal implications of the explicit mentioning of automation bias (AB) in the Artificial Intelligence Act (AIA). The AIA mandates human oversight for high-risk AI systems and requires providers to enable awareness of AB, i.e., the human tendency to over-rely on AI outputs. The paper analyses the embedding of this extra-juridical concept in the AIA, the asymmetric division of responsibility between AI providers and deployers for mitigating AB, and the challenges of legally enforcing this novel awareness requirement. The analysis shows that the AIA’s focus on providers does not adequately address design and context as causes of AB, and questions whether the AIA should directly regulate the risk of AB rather than just mandating awareness. As the AIA’s approach requires a balance between legal mandates and behavioural science, the paper proposes that harmonised standards should reference the state of research on AB and human-AI interaction, holding both providers and deployers accountable. Ultimately, further empirical research on human-AI interaction will be essential for effective safeguards.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper analyzes the AI Act’s explicit reference to automation bias (AB) in Article 14(4b), arguing that the provider-focused duty to enable awareness of AB does not match the deployer-controlled causal factors and is difficult to enforce legally. It proposes that harmonised standards reference the state of research on AB and human-AI interaction, with responsibilities for both providers and deployers.
Key findings
- The AIA mentions AB only in Article 14(4b), obliging providers to enable human oversight agents to “remain aware” of the tendency to over-rely on AI outputs; Article 26(2) imposes deployer duties on competence, training, and support but does not explicitly mention AB, though the paper argues deployers are plausibly obliged insofar as they control AB’s causal factors.
- AB is multifactorial: provider-controlled design/interface/instructions and deployer-controlled individual, motivational, organisational, and contextual factors both matter, so the AIA’s asymmetric provider focus does not adequately address design and context as causes of AB.
- Enforcement is challenging because “awareness” is a subjective internal state, a single decision rarely proves AB, ex post experimental re-enactments are likely prohibitively costly and time-consuming, and expert testimony is presented as the most probable proof route; the paper also notes that the unbiased “true outcome” is often unavailable, using a hiring example where a provider’s 80% selection-rate threshold flags female selection at 0.30 versus male at 0.40 (75% of the best-performing group) with ten male and ten female candidates and three female/four male invited.
- The paper identifies a protection gap in Article 6(3) AIA: its high-risk exception relies on provider self-assessment and does not reflect the AB concern, and it suggests requiring an explicit reference to Article 14(4b) safeguards in that self-assessment; it also notes the Commission assessed that 5–15% of AI systems on the EU market are high-risk.
Methods
- Legal-doctrinal analysis of the AIA, including Article 14(4b), Article 26(2), Article 6(3), legislative history, and the relationship with Article 22 GDPR and the CJEU’s SCHUFA decision.
- Draws on psychological and human-AI interaction literature on AB, cognitive biases, and algorithm aversion, plus illustrative examples such as the hiring/fairness scenario from Langer et al.
- Main analysis is normative and interpretive: it assesses the division of responsibility between providers and deployers, the justiciability of the awareness requirement, and proposes harmonised standards with a state-of-research clause.
Limitations
- Stated: the paper repeatedly notes scientific uncertainty about the causes, remedies, and interactions of AB in high-risk AI, that psychological research on human-AI interaction is complex and dynamic, and that legal scholars disagree on how law can reduce cognitive biases.
- Stated: enforcement limitations include difficulty proving AB occurred, the insufficiency of a single decision, costly ex post re-enactments, debate over the predictive value of the IAT, and unstable ground truth for fairness/non-discrimination and legal proceedings.
- Evident in the text: the analysis is conceptual/doctrinal rather than empirical, relies on secondary literature and examples, and notes that harmonised standards addressing AB have not yet been published.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
72 Sources not found · 66 Own interpretation · 10 No basis given · 1 Own result
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
“Our analysis of the AIA revealed that the obligation to enable awareness of AB can arguably be assigned to both the provider and the deployer of high-risk AI systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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: passes — The paper gives a reason why it holds.
In the paper
“Article 14 AIA prescribes human oversight to mitigate the risks of high-risk AI systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites Article 14 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“As stated in section III.1, organisational choices such as decision-makers' workload and their work environment have been shown to affect the emergence of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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: unclear — It is unclear why the paper says it holds.
In the paper
“Article 14(4b) AIA obliges AI providers to deliver their AI systems to AI deployers in a way that natural persons to whom human oversight is assigned, are enabled “to remain aware [emphasis added] of the possible tendency of automatically relying or over-relying on the output produced by an AI system (automation bias).””
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“These efforts demonstrate that debiasing through enabling awareness is not entirely unprecedented.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [30]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Through textual and systematic interpretation, it is thus at least possible (and we believe plausible) to assume that deployers have an obligation to enable awareness of AB insofar as they control the causal factors of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
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
“A failure to assess whether measures intended to raise awareness of bias actually achieved this goal renders the requirement for effective human oversight meaningless.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could 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 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
“Others, such as workload distribution, can lower the likelihood of AB without, however, necessarily raising human oversight agents' awareness of their own potential bias.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plaindirect- 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.
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
“Moreover, as mentioned in section III.1 above, the psychological literature understands AB primarily as a matter of attention, not of discriminatory association.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plain- 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: unclear — It is unclear why the paper says it holds.
In the paper
“If overcorrection becomes systematic, it could introduce a bias against AI outputs among those tasked with oversight (on algorithm aversion, see below, section V.1), 30”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [30]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Firstly, if AB occurs in human oversight, then it occurs at the level of the deployer, who in complying with the AIA, assigns oversight functions to natural persons who may be susceptible to AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
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: unclear — It is unclear why the paper says 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
“While a particular human oversight process may be experimentally tested for its susceptibility to AB, such ex post re-enactments are likely prohibitively costly and time-consuming.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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: unclear — It is unclear why the paper says 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
“Section IV shows the difficulties of legally enforcing the obligation of enabling awareness, as proving AB occurred in a given human oversight system is challenging, not least because identifying the unbiased true outcome for many oversight decisions it will be difficult (if not impossible).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says 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
“Expert testimony is therefore the most probable way in which the occurrence of AB would be proven in legal proceedings.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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: passes — The paper gives a 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
“If providers (and deployers) bear the responsibility for effectively enabling awareness of AB, then the occurrence of AB could serve as evidence of failure to meet this obligation.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
In the paper
“AB has been demonstrated in various applications 8 and is difficult to combat because it is multifactorial: technical, psychological, social and normative factors can all play a role and interact with each other.9”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [9]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“There is a considerable risk that providers will interpret their obligation under Article 14(4b) AIA as a simple notification requirement, informing human oversight agents about the existence and potential causes of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: unclear — It is unclear why the paper says it holds.
- Not matched
In the paper
“The obligation to “enable awareness” under Article 14(4b) AIA creates further legal uncertainty as regards its enforcement.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
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
“However, AB is not a social bias such as race or gender bias.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
In the paper
“To ensure compliance with Article 14(4b) AIA, AI providers may thus be incentivised to design their AI systems to encourage human oversight agents to overcorrect rather than undercorrect system outputs, for example through notifying the human oversight agent of their doubts in the reliability or correctness of the outcome.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“However, a single decision will rarely be sufficient to demonstrate that AB influenced the oversight outcome.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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: passes — The paper gives a reason why it holds.
- Rests on an assumption: Establishing the occurrence of psychological bias usually requires multiple data points.
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 AIA thus divides the responsibility for enabling effective human oversight between the AI provider, who is responsible for the development and design of the AI system, and the AI deployer, who is responsible for the organisational implementation of human oversight.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Despite its sweeping regulation of Artificial Intelligence (AI) across sectors, 1 the European Union's AI Act (AIA) leaves important normative decisions to AI developers and users. 2”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [1]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [2]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Factually, deployers thus indeed have measures within their control that can enable awareness of AB and reduce the likelihood of biased oversight decisions.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
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 AIA does not provide explicit guidance on how providers should ensure human oversight agents are made aware of their vulnerability to AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
In the paper
“The psychological literature understands AB primarily (albeit not exclusively) as a matter of attention and its (mis) placement.31”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [31]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- 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.
- Not matched
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
“Assuming that these factors also apply to AB for human oversight, it must be said that the provider cannot control for them all.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
- The authors state this as an assumption: The factors that cause automation bias generally also apply to automation bias in human oversight of high-risk AI systems.
In the paper
“This would, however, exceed the requirements of the wording of Article 14(4b) AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Deployers' human oversight agents thus seem to be responsible for monitoring the accuracy of AI systems measured against the accuracy expected by the providers.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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: passes — The paper gives a 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
“In turn, this significantly complicates the determination of whether a human oversight outcome has been unduly influenced by AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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
“As Langer et al. (who are drawing on the same example) show, a human oversight agent who detects a selection rate of female candidates of 0.30 and of male candidates of 0.40 would thus have numerical evidence of unfairness in the AI's outputs (as the minority rate is at 75% of the best-performing group)”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Langer et al.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedgedcorrelational- 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.
- Not matched
In the paper
“However, as Langer et al. show, the deployer may still judge the process to have been fair after all: they may not be responding to the numerical evidence as sufficient evidence for unfairness in the selection process.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Langer et al.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a reason why it holds.
- Not matched
In the paper
“The concern expressed in Article 14 AIA about AB is not reflected in these criteria, which extensively broadens the scope of Article 6(3) AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [52]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“For some of these conditions however, AB may pose a significant problem.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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 demonstrates how the ground-truth for fairness and non-discrimination can vary between providers and deployers as well as the EU legislature and judiciary.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could 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
“To protect the fundamental rights of the job candidates as Article 14 AIA demands, the human oversight agent would not have to adhere to the providers' statistical measures of fairness but instead consider the requirements of EU non-discrimination law.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a reason why it holds.
- Not matched
In the paper
“and AB is not the only cognitive bias relevant to human-AI interaction.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [51]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- 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.
- Not matched
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 potential drift of the ground truth reveals the instability of our understanding of the unbiased true outcome for an important subset of tasks of human oversight agents.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could 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
“However, Article 6(3) AIA introduces a substantive exception: systems usually referred to in Annex III (education, credit scoring, public services etc.) are not considered high-risk where there is no significant risk of harm to health, safety, or fundamental rights, as determined by a provider's self-assessment before the system is launched on the market.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 6(3) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“Moreover, legal scholars do not agree on how the law can be utilised to reduce cognitive biases.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [49]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“However, even if the system is meant to only “support” a human decision, AB may still cause humans to unduly over-rely on the AI's suggestion.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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: unclear — It is unclear why the paper says 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
“nor do they provide standards for verifying compliance”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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 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
“In sum, if providers and deployers are obliged to effectively raise awareness of AB and thus reduce the likelihood of AB, then legal enforcement of this obligation faces considerable challenges.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could 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: Raising awareness of AB reduces the likelihood of AB.
In the paper
“The fact that the exception in Article 6(3) AIA relies on self-assessment by the providers without ex ante review by supervisory authorities”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 6(3) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“but whether they will judge this output to actually reflect unfairness will often be left to the agents' discretion or that of their supervisor.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Langer et al.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a reason why it holds.
- Not matched
In the paper
“How the multitude of biases that potentially apply to a given context interact with one another has not been well understood”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [50]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“creates a notable protection gap”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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 current state of research on AB would likely reveal a significant degree of scientific uncertainty about causes, remedies, and interactions with other cognitive biases.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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: unclear — It is unclear why the paper says it holds.
In the paper
“Thus far, EU non-discrimination law has relied less on statistical evidence to prove prima facie discrimination, instead defining the requirements for legally classifying the treatment of a group as discriminatory by context and judicial intuition.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [46]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Article 14(4b) AIA does not address these concerns, as its application is contingent on the AI system being deemed high-risk, a premise that the exception under Article 6(3) AIA specifically negates.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 6(3) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“In issuing this criterion, the ECJ took a comprehensive view of the decision-making situation focusing on more than the formal responsibility of the person involved, to include (for this case) the entire process of granting a loan.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites ECJ C-634/21
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“Article 22 GDPR additionally provides exceptions for situations where the data subject is not to be protected from fully automated decisions.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 22 GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- 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.
- Not matched
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
“Provider and deployer remain two different data processors.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
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.
In the paper
“we argue that compliance with the (unclear) requirements of Articles 14(4b) and 26(2) AIA does not rule out an automatic decision under Article 22 GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites ECJ C-634/21
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Consequently, a decision may still constitute an automated decision under Article 22 GDPR even if the provider and deployer fulfil the requirements of Articles 14(4b), 26(2) AIA, and responsible persons are made aware of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 22 GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Articles 14(4b), 26(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“However, the provider also being subject to specific obligations under Article 14(4b) AIA to ensure human oversight and measures against AB, suggests a joint controllership between provider and deployer under Article 26 GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 26 GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedgedconditional- 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.
- Not matched
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“there is thus no joint controllership under Article 26 GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites ECJ C-683/21
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites ECJ C-604/22
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Although the regulations are fundamentally different, the requirement for human oversight in the AIA pursues a similar goal to that of Article 22 GDPR: affected parties should not be subject to automated decisions without the possibility of human intervention.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
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.
In the paper
“Responsibility for the processing of personal data by the deployer at the stage of actual application of an AI system is relatively clear: the deployer is the controller under Article 4(7) GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 4(7) GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boosted- 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.
- Not matched
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 responsibilities of the AIA are not the same as those of the GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
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.
In the paper
“In practice however, those who only provide the AI system do not have any influence on the specific data processing carried out by those deploying it.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites [61]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“according to the ECJ, the criterion for an automated decision under Article 22 GDPR is whether the subsequent final decision (in this specific case, the granting of the loan) was decisively based on the preceding automated decision, even if the responsible person has formal and substantive decisionmaking powers.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites ECJ C-634/21
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirectSingle case- 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.
- Not matched
In the paper
“The SCHUFA system itself is unlikely to fall under the AIA, as it is not an AI system under Article 3(1) AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 3(1) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedged- 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.
- Not matched
In the paper
“exact requirements for the level of human involvement with the machine decision proposal remain unclear”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites [58]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
- Probably not a substantial claim. 68
- It may state two separate claims as one. 27
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
“The AIA, however, only explicitly mentions AB in stating the obligations of providers.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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
“It also remains unclear what kind of measures providers (and deployers) should take to address AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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 have also shown how further empirical challenges await the enforcement of the enabling of awareness and require deference to experts in psychology.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
plain- 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.
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
“Our analysis of AB in the AIA raises critical questions about the justiciability of the requirement to enable human oversight agents' self-awareness of their susceptibility to AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
plain- 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.
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
“counteracting the AIA's goal of removing bias in human oversight.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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
“To the best of our knowledge, no legal obligations currently exist in the EU that require making an individual aware of their susceptibility to cognitive bias, rendering a comparative legal assessment within EU law impossible.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
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
“All these measures lie within the control of deployers (although some may prove to be costly to implement).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“It would then be sanctionable according to Article 99(4a) AIA in conjunction with Article 16(a) AIA and therefore subject to a substantial fine.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 99(4a) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 16(a) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“However, the psychological literature still debates the predictive value of the IAT 36 which potentially lowers the probative value of unconscious bias tests in legal procedures.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [36]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedgedconditional- 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.
- Not matched
In the paper
“Although discussions in the legislative process included shifting a significant aspect of oversight towards the users and deployers of AI systems, 28 this has not been reflected in the final wording of Article 14 AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
What the sources say
- Not matchedcites [28]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Measures to prevent AB also risk leading to overcorrection and humans overly discounting AI systems' outputs, potentially eroding the advantages of introducing AI systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
In the paper
“Article 26(2) AIA states that deployers “shall assign human oversight to natural persons who have the necessary competence, training, and authority, as well as the necessary support.””
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 26(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“This explicit recognition of the risk of AB is no longer found in the recitals of the AIA (see recital 73).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
What the sources say
- Not matchedcites recital 73
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“It is notable that the AIA does not consider the fact that systems themselves are particularly conducive to AB due to their design as a criterion for a high-risk classification.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
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.
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
““Awareness” is a subjective internal state.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
plain- 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
“Even more surprisingly, AB is the only psychological bias named by the AIA, despite human oversight agents likely being subject to other types of biases, too.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
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: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
In the paper
“Psychological research suggests that AB is not only caused by design features at the system level controlled by the provider, but also by individual, motivational, organisational, and contextual factors controlled by the deployer.29”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
What the sources say
- Not matchedcites [29]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“As such, AIA does not chart an entirely new path, although it appears to be the first EU regulation to demand consideration of concepts from behavioural research for its implementation.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
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: 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.
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
“These obstacles to empirical testing suggest that expert testimony based on a review of the oversight conditions may offer the more practical route.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
- Rests on an assumption: A legal procedure will typically focus on a specific case of oversight failure that may involve a single human decision or multiple human decisions.
- Rests on an assumption: Establishing the occurrence of psychological bias usually requires multiple data points.
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
“However, the provider is clearly able to control the design of the AI system and its user interface as well as the content of instructions (including information about potential susceptibility to AB) provided to the deployer.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
boosted- 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
“In the academic literature, automation bias (AB) describes the psychological phenomenon of human overreliance on automation in decision-making, i.e., the human tendency to overly accept the suggestions of an automated system.6”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [6]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“AB is one of many cognitive biases such as confirmation bias, availability bias, or algorithm aversion (see further below) that may lead to human error in the oversight of AI systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
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.
Probably not a substantial claim.
In the paper
“It is neither a new development of digitalisation or specific to AI, nor has it arisen during recent technical developments such as large language models (LLMs).Instead, it is a comparatively old phenomenon, well documented in other settings such as aviation.12”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [12]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
It may state two separate claims as one.
- Not matched
In the paper
“The norm does not explicitly mention AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 26(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“However, psychological research into human-AI interactions is complex, dynamic, and likely subject to a significant degree of scientific uncertainty about the causes of AB in high-risk AI applications.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
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: unclear — It is unclear why the paper says 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
“It argues that the causal factors of AB do not match the legal division of responsibilities and thus creates a tense asymmetry: while the provider is responsible for enabling awareness of AB, the bias manifests itself in the deployer's context.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
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: unclear — It is unclear why the paper says 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
“To our knowledge, Article 14 AIA establishes the first norm of EU law that explicitly mentions cognitive bias.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
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.
In the paper
“AB in particular is often understood as a lack (or misplacement) of attention.11”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [11]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“As the norm requires human oversight agents “remain aware” (Article 14(4)(b) AI Act) of their susceptibility to AB, proper adherence requires consideration of the current state of empirical AB findings in behavioural psychology.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
What the sources say
- Not matchedcites Article 14(4)(b) AI Act
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
- Not matched
In the paper
“Although AB is only one of many psychological phenomena relevant to human-AI interactions, with its inclusion, the AIA answered long-standing calls for greater recognition of behavioural factors in legislation and legal scholarship by considering psychological findings in legal compliance and adjudication.23”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
What the sources say
- Not matchedcites [23]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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 list of measures is not exhaustive, and some, such as education about the existence of bias, fall within the literal meaning of enabling awareness.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
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
“While this definition is legally expedient, it is also scientifically simplified, and does not reflect the decades of psychological research into AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Effective risk governance measures such as technical documentation (Article 11 AIA), record-keeping (Article 12 AIA), and transparency (Article 13 AIA) would thus be crucial to document oversight decisions for expert review.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 11 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 12 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 13 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedged- 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.
- Not matched
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
“It will be shown that the AIA's legislative history does not provide an answer as to why AB is the only cognitive bias that is mentioned in the legal text.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
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
“The AIA's focus on the provider is a consequence of the European approach to product safety law and its focus on the product providers.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
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: 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
“The reference to AB in Article 14(4)(b) AIA was incorporated into the Commission draft of 21 April 2021.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. Automation bias and AI
What the sources say
- Not matchedcites Article 14(4)(b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“Moreover, common debiasing strategies (beyond specifically considering AB) include measures such as educational training on the existence and occurrence of implicit biases, testing individuals' susceptibility to bias before selecting them for their role, and increasing diversity amongst teams.33”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [33]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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 singular focus on AB in Article 14 AIA may thus over-emphasise the risk of AB to the detriment of addressing other cognitive biases.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
In the paper
“AB has been demonstrated in various applications 8”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
What the sources say
- Not matchedcites [8]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boosted- 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.
- Not matched
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
“However, overcorrecting could be costly for the AI deployer, who will incur the error costs arising from false overcorrection, and for the subjects of the AI-driven decision, whose rights may be violated (assuming that overcorrection can pose risks to safety, health, and fundamental rights).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 1. AI providers' obligation to enable awareness of automation bias
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.
- The authors state this as an assumption: Overcorrection can pose risks to safety, health, and fundamental rights.
In the paper
“However, we reject this view, given Article 14 AIA explicitly requires human oversight to be effective, thereby establishing effectiveness as the standard for legal assessment.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boosteddirect- 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.
- Not matched
In the paper
“Article 22 GDPR, the prohibition of fully automated decisions when processing personal data, has so far overlooked the issue of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 22 GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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 anticipated effect of uncertainty is tighter judicial scrutiny of the credentials and provenance of the social scientists involved in AB research, with stronger weight attached to findings from public (or at least industry-independent) research institutions (see above, section IV.1).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
hedgeddirect- 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.
In the paper
“Algorithm aversion, for example, denotes humans' tendency to trust an algorithm's opinion less than that of a human even when the algorithm is shown to be more accurate.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [51]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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 fundamental mismatch between the European product safety regime and fundamental rights protection raises normative concerns.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
In the paper
“The technical documentation for AI systems demanded by the AIA includes information on “degrees of accuracy for specific persons or groups of persons,” the “overall expected level of accuracy” and “foreseeable unintended outcomes and sources of risks to health and safety, fundamental rights and discrimination” (No. 3 Annex IV of the AIA).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites No. 3 Annex IV of the AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“For example, Article 6(3)(c) AIA excludes an AI system from the high-risk category if the system “is not meant to replace or influence the previously completed human assessment, without proper human review.””
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 6(3)(c) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“Article 6 AIA defines the classification methodology for high-risk systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 6 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“but such an obligation would be legally and logically consistent.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
hedged- 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.
In the paper
“State-of-research or state-oftechnology clauses are common in regulatory law and could directly be included in the wording of Article 14 AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
What the sources say
- Not matchedcites [56]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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. It may state two separate claims as one.
- Not matched
In the paper
“Article 11 AIA suggests that AI providers inform deployers about how their AI systems should perform.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites No. 3 Annex IV of the AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Not matched
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
“The AIA merges two fundamentally different regulatory approaches: product safety law and the protection of fundamental rights.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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.
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 proposed state-of-research clause would at least open the legal implementation of the AIA to incorporating findings from social science.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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.
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
“Harmonised standards provide an alternative forum for addressing AB under the AIA, offering more room for extra-juridical factors to be included in the standardised text.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
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
“It is also unclear how to prove a human decision was not influenced by the AI system.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
plain- 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
“Similarly, not all measures which could be taken to reduce cognitive bias necessarily raise awareness of a susceptibility to bias.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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 legal uncertainty about how to reduce AB in human oversight is inherently intertwined with the challenging regulatory structure of the AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
plaincorrelational- 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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Moreover, in legal proceedings (a high-risk domain for AI systems according to the AIA), there is no procedure-independent ground-truth available at all: the only criterion for assessing the outcome of a legal procedure is another legal procedure.”
- Also stated in: § 2 AI deployers' obligation to enable awareness of automation bias · plain
“There are no criteria external to the legal procedure that can establish the correctness of a judicial decision.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites [47]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [48]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could 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.
- Also stated in: § 2 AI deployers' obligation to enable awareness of automation bias · plain
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“A system can be classified as high risk either because it is a product or a safety component under the NLF (Article 6(1) AIA) or because it falls under the use-cases of Annex III (Article 6(2) AIA).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites Article 6(1) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 6(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“References to AB make these already complex relationships even more complicated by explicitly referring to extra-juridical factors while also requiring providers to enable “proportionate” (Article 14(4) AIA) awareness of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
What the sources say
- Not matchedcites Article 14(4) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“the psychological literature on AB in human-AI interaction is still emerging, with commensurate scientific uncertainty about the causes of (and remedies for) AB in high-risk AI use cases.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
What the sources say
- Not matchedcites [49]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [50]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“The AIA applies in principle alongside the GDPR and does not affect its scope of application (Article 2(7) AIA).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 2(7) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- 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.
- Not matched
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
“an amendment of the AIA's text is rather unlikely to happen anytime soon, whereas standards are still being drafted.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
hedged- 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. 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 criteria do not define standards for the role and responsibilities of the human making the decision”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
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 state-of-research clause would thus require providers (and, as we argue, ideally also deployers) to continuously update and verify their knowledge on the causes of and remedies to AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
hedgeddirect- 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.
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 AIA thus unnecessarily limits the scope of debiasing measures for AB to awareness-raising measures.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
In the paper
“The legal definitions in the AIA embrace technical developments in the field of AI and transfer them into the normative system of law (see section II.2).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
What the sources say
- Not matchedcites section II.2
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“It appears that for the time being, expert testimony is the only viable method for proving AB has occurred or that a given human oversight scheme is prone to AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
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.
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 AIA circumnavigates the problem of detailing an under-studied psychological phenomenon at the cost of legal uncertainty about how to enforce the enabling of AB awareness (see section IV above).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > V. Discussion
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.
In the paper
“It may be captured by industry interests and again raises the question of whether standards should entail detailed rules or flexible principles.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
What the sources say
- Not matchedcites [57]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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: passes — The paper gives a 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.
- Not matched
In the paper
“However, according to Article 14 AIA, the scope of human oversight clearly goes beyond errors of accuracy and includes the protection of fundamental rights.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
What the sources say
- Not matchedcites Article 14 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boosted- 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.
- Not matched
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
“The latter relies on proportionality assessments to provide the greatest possible protection from risk for an individual's rights.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2. Harmonised standards should reference the state of research
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: 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.
In the paper
“The joint EDPB-EDPS opinion on the AIA emphasises the intersection of the provisions and states that qualified human oversight could ensure that the Article 22 GDPR requirement -“the right not to be subject to a decision based solely on automated processing”is respected.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites EDPB-EDPS Joint Opinion 5/2021
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
Probably not a substantial claim.
- Not matched
In the paper
“If awareness alone proves to be insufficient, this article argued that the legal obligation under Article 14 AIA for human oversight to be effective would require AI providers (and deployers) to adopt proportionate and evidence-based debiasing interventions aligned with the most recent and state of the art research on AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
What the sources say
- Not matchedcites Article 14 AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
In the paper
“For now, the implementation of Article 14(4b) AIA must navigate a considerable degree of both scientific and normative uncertainty.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- 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.
- Not matched
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 would also require aligning the currently asymmetric division of responsibilities between AI providers and deployers for AB awareness with their actual influence on the causal factors of AB.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
hedgedcorrelational- 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.
In the paper
“According to paragraph 2(a), the prohibition does not apply if the decision is necessary for entering into or performing a contract, (b) the decision is authorised by Union or national law and provides for adequate safeguards for the rights and legitimate interests of the data subject, or (c) the decision is based on consent.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 22(2) GDPR
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“Moreover, it remains uncertain whether the legislator considers mere awareness of AB to be an adequate remedy or whether additional measures are required to mitigate its effects.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
plain- 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
“Furthermore, it remains unclear why AB is the only cognitive bias explicitly mentioned in the legal text of the AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
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: 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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“Therefore, it remains unclear how a precise distinction of responsibility could be made either in terms of a breach of Article 14(4b) AIA or Article 26(2) AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 26(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
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.
- Not matched
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
“externally proving a decision was actually made by a human is almost impossible.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
plain- 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
“However, these exceptions apply to situations in which the interests of the data subjects are otherwise protected, either through voluntary entry into a contract, a decision by the legislator, or through their own consent.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
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.
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 an obligation would go beyond fostering awareness, demanding active engagement with strategies to counteract bias at the design and implementation stages of AI systems.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
hedgeddirect- 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.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“This would also be in line with the complementary obligations under Article 14(4b) and Article 26(2) AIA.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
What the sources say
- Not matchedcites Article 14(4b) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Article 26(2) AIA
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
hedgedcorrelational- 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.
- Not matched
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 focus on AB alone may suggest a prioritisation of human autonomy: the AIA deems human oversight agents erroneously rejecting an AI's recommendation as less problematic than human oversight agents erroneously following an AI's recommendation.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
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. It may state two separate claims as one.
The authors state this as an assumption.
High-risk AI systems will have great practical relevance for the European single market.
“It is reasonable to assume that high-risk systems will have great practical relevance for the European single market.”
The authors state this as an assumption.
The factors that cause automation bias generally also apply to automation bias in human oversight of high-risk AI systems.
“Assuming that these factors also apply to AB for human oversight, it must be said that the provider cannot control for them all.”
The authors state this as an assumption.
Overcorrection can pose risks to safety, health, and fundamental rights.
“(assuming that overcorrection can pose risks to safety, health, and fundamental rights)”
The paper relies on this without support.
A legal procedure will typically focus on a specific case of oversight failure that may involve a single human decision or multiple human decisions.
“A legal procedure will typically focus on a specific case of oversight failure that may involve a single human decision or multiple human decisions.”
The paper relies on this without support.
Establishing the occurrence of psychological bias usually requires multiple data points.
“Establishing the occurrence of psychological bias usually requires multiple data points.”
The paper relies on this without support.
Raising awareness of AB reduces the likelihood of AB.
“if providers and deployers are obliged to effectively raise awareness of AB and thus reduce the likelihood of AB”
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
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > I. Introduction
“Article 22 of the General Data Protection Directive (GDPR).”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > VI. Conclusion
“The General Data Protection Regulation and Automated Decision-Making: Will It Deliver?”
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
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 2 AI deployers' obligation to enable awareness of automation bias
“Arguably, the wording of Article 14(4b) AIA only states that deployers must enable human oversight agents' awareness of their susceptibility to AB”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > III. Who is responsible for de-biasing human oversight?
“However, the obligation to enable awareness of AB is limited to providers in Article 14 AIA.”
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
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
“the deployer is the controller under Article 4(7) GDPR.”
§ Automation Bias in the AI Act: On the Legal Implications of Attempting to De-Bias Human Oversight of AI > 3. Relationship with the GDPR
“Provider and deployer remain two different data processors.”
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 5
- JJ Rachlinski and SL Johnson, "Does Unconscious Racial Bias Affect Trial Judges" (2009) 84 Notre Dame Law Review 1195.
- J Agerström and D-O Rooth, "The Role of Automatic Obesity Stereotypes in Real Hiring Discrimination" (2011) 96 Journal of Applied Psychology 790; FL Oswald and Others, "Predicting Ethnic and Racial Discrimination: A Meta- Analysis of IAT Criterion Studies" (2013) 105 Journal of Personality and Social Psychology 171.
- T Tridimas, The General Principles of EU Law (2. ed., 1. publ. in paperback, Oxford, Oxford University Press 2007) p 216. 38 Tridimas (supra note 35) 216. 39 Laux (n 20) 240-64. The EU AI Office could play an important role in monitoring the field of psychological research on human-AI interaction, see H Pouget and J Laux, "A Letter to the EU's Future AI Office" (3 October 2023) available at <https://carnegieendowment.org/2023/10/03/letter-to-eu-s-future-ai-office-pub-90683> (last accessed 25 January 2024). 40 AIA Annex IV 3.
- M Langer, K Baum and N Schlicker, "Effective Human Oversight of AI-Based Systems: A Signal Detection Perspective on the Detection of Inaccurate and Unfair Outputs" (2024) 35 Minds and Machines 1. 42 This 80% rule is taken from the example in: L Alexander, Making Decisions about Adverse Impact: The Influence of Individual and Situational Differences (Rice University 2022) available at <https://hdl.handle.net/1911/114147> accessed 15 January 2025. Cited after: Langer, Baum and Schlicker (supra note 39). 43 Langer, Baum and Schlicker (supra note 39) 12-13. 44 Ibid., 13. 45
- Wachter, Mittelstadt and Russell (supra note 14). 46 Ibid. 47 On this problem and jury decisions, see: J Elster, Securities against Misrule: Juries, Assemblies, Elections (New York, Cambridge University Press 2013). For an application to the EU legal system, see: Laux (n 20). For the context of human oversight, see: Langer, Baum and Schlicker (supra note 39) 12. 48 JA Jolowicz, On Civil Procedure (Cambridge, Cambridge University Press 2000) p 86.
Cited by
Sorted by citations
Štrukelj, T., & Dankova, P. (2025) · Administrative Sciences · doi:10.3390/admsci15070274 · Cited by 30
On the Complexities of Testing for Compliance with Human Oversight Requirements in AI Regulation
Langer, M., Lazar, V., & Baum, K. L. (2025) · Lecture notes in computer science · doi:10.1007/978-3-032-07132-3_11 · Cited by 12
From European Standards to Responsible Use of AI in Clinical Practice
Jessen, R., Bak, M., Muller, C. C. J., Verkooijen, H. M., Yeung, K. K., & Ploem, M. C. (2025) · European Journal of Health Law · doi:10.1163/15718093-bja10143 · Cited by 4
Maska, L., Vlachopanou, P., Kalamaras, D., & Tsameti, A. (2025) · Research Square · doi:10.21203/rs.3.rs-7665675/v1 · Cited by 2
Maryahin, R. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6539278
Metadata
- Authors
- Laux, J., & Ruschemeier, H.
- Affiliations
- University of Oxford (GB)
- FernUniversität in Hagen (DE)
- Corresponding author
- Laux, Johann
- Venue
- European Journal of Risk Regulation
- Publisher
- Cambridge University Press (CUP)
- Published
- 2025-07-28
- Volume
- 16
- Issue
- 4
- Pages
- 1519-1534
- ISSN
- 1867-299X (ISSN-L)
- 2190-8249
- OpenAlex
- W4412687403
- Semantic Scholar
- 3041ddab9dc1483c2be3903df2bbf6bdec7ba691
- Cited by
- 69
- References
- 0
- Citations by year
- 2026: 58 · 2025: 11
- FWCI
- 149.60
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- European Journal of Risk Regulation (publishedVersion, cc-by)
- SSRN Electronic Journal (submittedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Social Sciences › Law › Digitalization, Law, and Regulation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Law › Artificial Intelligence in Law
- Keywords
- automation bias, EU AI Act, human oversight of AI, high-risk AI systems, human-AI interaction, harmonised standards, AI regulation
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 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (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
