Pith. sign in

REVIEW 3 major objections 5 minor 36 references

WebXAII: an open-source web framework to study human-XAI interaction

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read WebXAII lets researchers define complete human-XAI experiments in JSON configuration files that can be shared for replication.

desk verdict A useful, honestly-scoped open-source framework for human-XAI experiments; the 'full protocols' claim outruns the linear, precomputed architecture, but the artifact and the published-protocol reproduction are real. read the letter →

arxiv 2506.14777 v2 pith:NP6X4HJR submitted 2025-05-16 cs.HC cs.AIcs.LG

Jules Leguy , Pierre-Antoine Jean , Felipe Torres Figueroa , Sébastien Harispe This is my paper · ORCID
classification cs.HCcs.AIcs.LG
keywords XAIhuman-XAIinteractionexperimentalprotocolwebframeworkreproducibilityopen-sourcesoftwareJSONconfigurationexplainabilityevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper presents WebXAII, an open-source web framework that lets researchers run complete human experiments on explainable AI. Its central claim is that an entire experimental protocol—participant login, instructions, questionnaires, task-level decisions on AI predictions and explanations, feedback, and data logging—can be expressed as a JSON configuration file and deployed without writing bespoke interface code. The paper argues this matters because human-XAI studies usually rely on one-off interfaces that are not shared, which slows replication and reuse. It supports the claim by reproducing, in WebXAII, the interface and protocol structure of a published state-of-the-art study on AI overreliance, using only configuration files.

What carries the argument

The central mechanism is a composite, view-driven architecture defined entirely in JSON. Four modules—Connection, Protocol, Experiment, and Task—manage successive stages of a participant's journey, while four views—Questionnaire, Instruction, Instance decision, and Feedback—handle the concrete interactions, with the Instance decision view as the fundamental unit where participants act on a model prediction and its explanations. The architecture works because a protocol is a declarative document, not a program: nesting modules, randomizing instance order within tasks, and binding participant groups to different protocol configurations all happen through the JSON description. Its boundary condition is equally important: because AI and XAI outputs are precomputed externally, the framework never needs to run a model during the experiment.

What would settle it

Try to encode the reproduced study's fifth experiment in WebXAII: participants choose whether to see an explanation, and that choice alters both the task and the reward. If no JSON configuration can represent this conditional flow—as the paper states in Section 5—then the framework's claim of embodying full protocols does not extend to choice-dependent designs; a successful implementation after a tree-based protocol extension would be the decisive countertest.

Watch

Extended reading notes

Core claim

The paper's central claim is that the visible surface of human-XAI studies decomposes into a small set of reusable building blocks: modules for connection, protocol, experiment, and task, plus views for questionnaires, instructions, instance decisions, and feedback. Arranging these blocks in the nested order Protocol > Experiment > Task > Instance decision is sufficient to represent a representative published protocol, including a between-group design with four experimental arms, which the implementation captures as four distinct protocols sharing the same structure. The framework deliberately keeps all machine-learning predictions and XAI explanations precomputed, presenting them as text or images rather than calling live models. The authors state this as an explicit design assumption in Section 3.1 and show through the reproduction that a rigorous modern study fits inside that boundary.

Load-bearing premise

The load-bearing assumption is that every implementable human-XAI protocol fits a linear Protocol > Experiment > Task > Instance decision structure with all AI and XAI outputs precomputed, which excludes protocols in which a participant's earlier choice changes what comes later, such as the reproduced study's fifth experiment that the paper concedes it cannot implement.

Editorial extensions

If this is right

  • A study's complete experimental interface can be published as one JSON file alongside its results, allowing others to replay the exact participant experience.
  • Multi-arm studies are handled by defining several protocol configurations and associating each participant group to one through the connection module, with no per-arm code.
  • Since all AI and XAI outputs are precomputed, deploying a study does not require hosting live machine-learning models during data collection.
  • Adding a new kind of question or display can be done as a new generic view without redesigning the protocol layer, a path the paper lists as planned future work.
  • Protocols where a participant's choice changes what comes later, such as the fifth study of the reproduced work, currently fall outside the framework's capacities.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: if WebXAII is adopted by multiple groups, shared JSON protocols could become a de facto standard for reporting human-XAI experiments, making replication cheap enough to become routine rather than rare.
  • Editorial inference: the framework's modules are not intrinsically tied to XAI, so the same JSON-protocol pattern could generalize to any human-subject study that presents precomputed stimuli and records choices, such as perception or judgment experiments.
  • Editorial inference: the paper demonstrates one representative protocol, not coverage across the field; a natural next test would be to encode a sample of recently published human-XAI protocols and measure what fraction fit the linear, precomputed structure without extension.
  • Editorial inference: if the list-based protocol structure is replaced by the tree-based structure the paper proposes, branching designs now excluded would become representable, and the framework would cover choice-dependent experiments such as the unreproduced fifth study.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces WebXAII, an open-source web framework for implementing interfaces for human-XAI interaction studies. The framework defines experimental protocols as JSON configurations composed of hierarchical modules (Connection, Protocol, Experiment, Task) and views (Questionnaire, Instruction, Instance decision, Feedback). The authors claim that full experimental protocols can be embodied without bespoke programming, and they validate this by re-implementing the protocol of Study 1 of Vasconcelos et al. (2023), a maze-based human-XAI study, using configuration files and screenshots. The paper also reviews related tools such as z-Tree, oTree, LimeSurvey, and ARDAS, and discusses planned extensions including dynamic model interactions and tree-structured protocols.

Significance. If validated, WebXAII addresses a real reproducibility gap: human-XAI study interfaces are typically bespoke and rarely shared. The contribution is concrete and checkable: the code and configuration files are publicly available, the architecture is clearly described, and the case study shows that at least one published protocol's structure can be represented with generic views. The JSON-configurable design is a useful step toward shareable, reproducible study interfaces. The main weakness is that the demonstration stops at static screenshots and generated stimuli rather than an end-to-end participant session, so the central capability claim is only partially evidenced. The paper honestly acknowledges the main architectural limitation, but that limitation conflicts with the unqualified 'full experimental protocols' claim in the abstract.

major comments (3)
  1. [Abstract; Section 3.1; Section 5] The abstract and Section 6 assert that WebXAII 'can embody full experimental protocols,' but Section 3.1 constrains protocols to a linear Protocol > Experiment > Task > Instance-decision structure with all AI and XAI outputs precomputed (assumptions 1-4), and Section 5 explicitly concedes that protocols in which participant choices affect later stages, such as Vasconcelos et al.'s Study 5, cannot be implemented. The single case study reproduces only Study 1, which lies entirely inside the supported class, so the boundary of the claimed generality is never tested. The 'full' claim should be qualified to the supported class, or additional cases that exercise the boundary should be demonstrated.
  2. [Section 4.2] The validation does not demonstrate an end-to-end participant-facing run. Section 4.2 states that the authors did not have access to the original maze stimuli and 'only use a small sample of maze images which we generated,' and the evidence presented consists of screenshots (Figures 3 and 4) and configuration files. No deployed session with participants is reported, and there is no verification that responses are logged, that timing constraints fire, or that randomization works as specified. Since the central claim is that the framework can effectively embody protocols, the authors should either report a deployed test session (even a pilot with a few users) or explicitly narrow the claim to structural reproduction.
  3. [Section 4.1; Section 3.1, Connection module] The implementation of Vasconcelos et al.'s Study 1 requires assigning participants to four experimental conditions (Protocols A-D), but Section 3.1 says only that 'each user identifier is associated with a specific protocol beforehand' and the case study does not describe how random assignment to conditions is realized. If participant-to-condition randomization is performed outside WebXAII, this should be stated; if WebXAII is expected to support it, this is a missing feature that affects the claim of embodying the protocol.
minor comments (5)
  1. [Throughout] The manuscript contains small language and formatting issues: 'javascript' should be 'JavaScript', 'configurations files' should be 'configuration files', and 'open-source software' should be 'open-source software' where it appears with the indefinite article.
  2. [Figure 2 caption] The caption uses 'Instructions View' while the body text and Figure 1 use 'Instruction view'; please standardize the terminology.
  3. [Section 4.2] The sentence 'we implement the structure of the study, but we only use a small sample of maze images which we generated' is an important limitation and should be reflected in the abstract or conclusion, not only inside the case-study section.
  4. [Section 3.1] The phrase 'which can then be transparent to the end-participant' is ambiguous; consider replacing 'transparent' with 'invisible' or 'not directly visible.'
  5. [References] Several DOIs have inconsistent spacing or formatting (e.g., references [15], [20], [21], and [23]); these should be normalized according to the journal style.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: WebXAII's central claim is validated against an external published protocol, and its acknowledged Study 5 limitation is a scope concern rather than a circular derivation.

full rationale

WebXAII is a software systems paper rather than a paper that derives a quantitative prediction from fitted inputs. Its central claim is that the framework can embody full experimental protocols, and this is demonstrated by reproducing the protocol of an external published study, Vasconcelos et al. [35]. The framework's architecture is defined in Section 3.1 as a Protocol > Experiment > Task > Instance-decision hierarchy with precomputed AI and XAI outputs, and Section 4 instantiates that architecture for Vasconcelos et al.'s Study 1. The benchmark is external, the configuration files and source code are publicly available on GitHub, and no load-bearing claim is justified by a self-citation chain. The only self-referential aspect is the generic one shared by all framework papers: the framework's own component structure is used to encode the target protocol. That is an implementation choice, not a derivation of the conclusion from itself. The paper explicitly concedes in Section 5 that Vasconcelos et al.'s Study 5, which requires participant choices to influence later protocol stages, cannot currently be implemented. This weakens the unqualified wording 'full experimental protocols' in the abstract, but that is an overclaim or scope issue, not circularity. No specific reduction of a claimed capability to an input, no fitted parameter renamed as a prediction, no imported uniqueness theorem, and no ansatz smuggled in via citation could be identified. Under the hard rule requiring a quotable reduction to establish circularity, the appropriate finding is no significant circularity, with score 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 1 invented entities

No data were fitted and no free parameters appear in this software paper. The central claim depends on three domain assumptions, all acknowledged in the text: the generality of the view hierarchy, the sufficiency of precomputed AI outputs, and the representativeness of the reproduced study. The main unvalidated assumption is the generality of the protocol abstraction beyond the single reproduced protocol.

assumptions (3)
  • domain assumption The Protocol > Experiment > Task > Instance decision hierarchy is sufficient to represent human-XAI experimental protocols.
    Assumed in Section 3.1 where the framework's structure is introduced. Section 5 acknowledges that adaptive or branching protocols, such as Vasconcelos et al.'s study 5, do not fit this structure.
  • domain assumption All AI and XAI computations can be performed before the experiment runs.
    Explicit assumption 3 in Section 3.1 states the framework displays precomputed outputs. Section 5 lists dynamic API interactions, for example with a language model, as future work, which means this assumption limits the current scope.
  • ad hoc to paper Vasconcelos et al.'s first study is representative of state-of-the-art human-XAI protocols.
    Section 4 states the study was chosen because it is 'complete and rigorous' and representative. This justifies generalizing from a single case study to the broader claim about the framework's flexibility.
invented entities (1)
  • WebXAII framework independent evidence
    purpose: A reusable, configuration-driven web framework for implementing whole human-XAI experiment protocols and collecting participant responses.
    The framework is a constructed software artifact with a public GitHub repository, screenshots, and configuration files, so a third party can deploy it and test the capability claim. This is not a purely theoretical construct.

how reviews work

0 comments
Cite this review

Pith. "Pith review of WebXAII: an open-source web framework to study human-XAI interaction." pith.science (2026). https://pith.science/paper/NP6X4HJR

@misc{pith2026250614777,
  author       = {Pith},
  title        = {Pith review of: WebXAII: an open-source web framework to study human-XAI interaction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NP6X4HJR}},
  note         = {Machine review of arXiv:2506.14777}
}
read the original abstract

This article introduces WebXAII, an open-source web framework designed to facilitate research on human interaction with eXplainable Artificial Intelligence (XAI) systems. The field of XAI is rapidly expanding, driven by the growing societal implications of the widespread adoption of AI (and in particular machine learning) across diverse applications. Researchers who study the interaction between humans and XAI techniques typically develop ad hoc interfaces in order to conduct their studies. These interfaces are usually not shared alongside the results of the studies, which limits their reusability and the reproducibility of experiments. In response, we design and implement WebXAII, a web-based platform that can embody full experimental protocols, meaning that it can present all aspects of the experiment to human participants and record their responses. The experimental protocols are translated into a composite architecture of generic views and modules, which offers a lot of flexibility. The architecture is defined in a structured configuration file, so that protocols can be implemented with minimal programming skills. We demonstrate that WebXAII can effectively embody relevant protocols, by reproducing the protocol of a state-of-the-art study of the literature.

Figures

Figures reproduced from arXiv: 2506.14777 by the authors.

Figure 1
Figure 1. Left : compositional representation of the main components of the WebXAII framework. The blocks outlined in dotted lines are optional. Alternative settings can be defined considering the same modules and views. Right : schematic representation of a possible Instance decision view, and JSON snippet to encode this view. 3. The WebXAII Framework: A Functional Overview WebXAII is a modular web framework developed in jav… view at source ↗
Figure 2
Figure 2. Schematic representation of the implementation of the study with WebXAII. Each column represents a protocol. The protocols differ from each other in the conditions of the maze solving task. The conditions are specified below each protocol. QV : Questionnaire View; IV : Instructions View. participants must solve the maze by themselves, without any help from the AI. They are informed immediately if they took the right… view at source ↗
Figure 3
Figure 3. Sample of screenshots of the interface. (a) Instruction view describing the first training experiment. (b) Instance decision view for the first training experiment (AI/XAI condition not shown) in the conditions of the protocol A or B (AI or XAI; Difficult) after submitting the decision (feedback displayed to the participant). (c) Instance decision view for the experimental phase in the conditions of the protocol D (… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Screenshot of the Questionnaire view for the survey at the end of the experiment. Currently, all interactions with machine learning and XAI techniques must be precomputed prior to the experiments. In the future, some of these interactions could be done dynamically via …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 9 canonical work pages

  1. [1]

    Hassija, V

    V. Hassija, V. Chamola, A. Mahapatra, A. Singal, D. Goel, K. Huang, S. Scardapane, I. Spinelli, M. Mahmud, A. Hussain, Interpreting Black-Box Models: A Review on Explainable Artificial Intelligence, Cognitive Computation (2024). doi: 10.1007/s12559-023-10179-8

    doi:10.1007/s12559-023-10179-8 2024

  2. [2]

    D. Minh, H. X. Wang, Y. F. Li, T. N. Nguyen, Explainable artificial intelligence: a comprehensive review, Artificial Intelligence Review (2022). doi:10.1007/s10462-021-10088-y

    doi:10.1007/s10462-021-10088-y 2022

  3. [3]

    Bylinskii, L

    Z. Bylinskii, L. Herman, A. Hertzmann, S. Hutka, Y. Zhang, Towards Better User Studies in Computer Graphics and Vision, Foundations and Trends in Computer Graphics and Vision (2023). doi:10.1561/0600000106

    work page doi:10.1561/0600000106 2023

  4. [4]

    R. Fok, D. S. Weld, In search of verifiability: Explanations rarely enable complementary performance in AI-advised decision making, AI Magazine (2024). doi: 10.1002/aaai.12182

    doi:10.1002/aaai.12182 2024

  5. [5]

    Šarčević, D

    A. Šarčević, D. Pintar, M. Vranić, A. Krajna, Cybersecurity Knowledge Extraction Using XAI, Applied Sciences (2022). doi:10.3390/app12178669

    work page doi:10.3390/app12178669 2022

  6. [6]

    Fresz, E

    B. Fresz, E. Dubovitskaya, D. Brajovic, M. F. Huber, C. Horz, How Should AI Decisions Be Explained? Requirements for Explanations from the Perspective of European Law, Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society (2024). doi:10.1609/aies.v7i1.31648

    doi:10.1609/aies.v7i1.31648 2024

  7. [7]

    Kloker, J

    A. Kloker, J. Fleiß, C. Koeth, T. Kloiber, P. Ratheiser, S. Thalmann, Caution or Trust in AI? How to design XAI in sensitive Use Cases?, AMCIS 2022 Proceedings (2022). URL: https://aisel.aisnet.org/ amcis2022/sig_dsa/sig_dsa/16

    work page 2022

  8. [8]

    R. Müller, How Explainable AI Affects Human Performance: A Systematic Review of the Be- havioural Consequences of Saliency Maps, International Journal of Human–Computer Interaction (2025). doi:10.1080/10447318.2024.2381929

    arXiv 2025

Show all 36 references
  1. [9]

    Z. C. Lipton, The mythos of model interpretability: In machine learning, the concept of inter- pretability is both important and slippery., Queue (2018). doi:10.48550/arXiv.1606.03490

    2018 doi arXiv

  2. [10]

    Zhang, P

    Y. Zhang, P. Tiňo, A. Leonardis, K. Tang, A survey on neural network interpretability, IEEE Transactions on Emerging Topics in Computational Intelligence (2021) 726–742. doi:10.1109/ TETCI.2021.3100641

    2021

  3. [11]

    Simonyan, A

    K. Simonyan, A. Vedaldi, A. Zisserman, Deep inside convolutional networks: Visualising image classification models and saliency maps, in: Workshop at International Conference on Learning Representations, 2014. doi:10.48550/arXiv.1312.6034

    2014 doi arXiv

  4. [12]

    J. T. Springenberg, A. Dosovitskiy, T. Brox, M. A. Riedmiller, Striving for simplicity: The all convolutional net, in: 3rd International Conference on Learning Representations, ICLR 2015, Workshop Track Proceedings, 2015. doi:10.48550/arXiv.1412.6806

    2015 doi arXiv

  5. [13]

    Smilkov, N

    D. Smilkov, N. Thorat, B. Kim, F. Viégas, M. Wattenberg, Smoothgrad: removing noise by adding noise, arXiv (2017). doi: 10.48550/arXiv.1706.03825

    2017 doi arXiv

  6. [14]

    Sundararajan, A

    M. Sundararajan, A. Taly, Q. Yan, Axiomatic attribution for deep networks, in: Proceedings of the 34th International Conference on Machine Learning, 2017. doi:10.48550/arXiv.1703.01365

    2017 doi arXiv

  7. [15]

    S. Bach, A. Binder, G. Montavon, F. Klauschen, K.-R. Müller, W. Samek, On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation, PloS ONE (2015). doi: 10. 1371/journal.pone.0130140

    2015

  8. [16]

    Achtibat, M

    R. Achtibat, M. Dreyer, I. Eisenbraun, S. Bosse, T. Wiegand, W. Samek, S. Lapuschkin, From attribution maps to human-understandable explanations through concept relevance propagation, Nature Machine Intelligence (2023). doi:10.1038/S42256-023-00711-8

    2023 doi

  9. [17]

    B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, A. Torralba, Learning deep features for discriminative localization, in: CVPR, 2016. doi: 10.48550/arXiv.1512.04150

    2016 doi arXiv

  10. [18]

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization, International Journal of Computer Vision (2020). doi:10.1007/s11263-019-01228-7

    2020 doi

  11. [19]

    Chattopadhay, A

    A. Chattopadhay, A. Sarkar, P. Howlader, V. N. Balasubramanian, Grad-CAM++: Generalized gradient-based visual explanations for deep convolutional networks, in: WACV, 2018. doi: 10. 1109/WACV.2018.00097

    2018

  12. [20]

    H. Wang, Z. Wang, M. Du, F. Yang, Z. Zhang, S. Ding, P. Mardziel, X. Hu, Score-cam: Score-weighted visual explanations for convolutional neural networks, in: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR Workshops, 2020. doi:10.48550/arXiv.1910.01279

    2020 doi arXiv

  13. [21]

    Zhang, F

    H. Zhang, F. Torres, R. Sicre, Y. Avrithis, S. Ayache, Opti-CAM: Optimizing saliency maps for interpretability, Computer Vision and Image Understanding (2024). doi:10.1016/j.cviu.2024. 104101

    2024 doi

  14. [22]

    why should i trust you?

    M. T. Ribeiro, S. Singh, C. Guestrin, "why should i trust you?": Explaining the predictions of any classifier, in: SIGKDD, 2016. doi:10.48550/arXiv.1602.04938

    2016 doi arXiv

  15. [23]

    S. M. Lundberg, S.-I. Lee, A unified approach to interpreting model predictions, in: NeurIPS, 2017. doi:10.48550/arXiv.1705.07874

    2017 doi arXiv

  16. [24]

    Vitali, D

    P. Vitali, D. Abir, S. Kate, Rise: Randomized input sampling for explanation of black-box models, BMVC (2018). doi:10.48550/arXiv.1806.07421

    2018 doi arXiv

  17. [25]

    Abnar, W

    S. Abnar, W. Zuidema, Quantifying attention flow in transformers, in: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Association for Computational Linguistics, Online, 2020. doi:10.48550/arXiv.2005.00928

    2020 doi arXiv

  18. [26]

    Chefer, S

    H. Chefer, S. Gur, L. Wolf, Transformer interpretability beyond attention visualization, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021. doi:10.48550/arXiv.2012.09838

    2021 doi arXiv

  19. [27]

    J. Wu, B. Duan, W. Kang, H. Tang, Y. Yan, Token transformation matters: Towards faithful post-hoc explanation for vision transformer, in: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. doi:10.48550/arXiv.2403.14552

    2024 doi arXiv

  20. [28]

    Chromik, M

    M. Chromik, M. Schuessler, A Taxonomy for Human Subject Evaluation of Black-Box Explanations in XAI., ExSS-ATEC@IUI (2020). URL: https://ceur-ws.org/Vol-2582/paper9.pdf

    2020

  21. [29]

    van der Waa, E

    J. van der Waa, E. Nieuwburg, A. Cremers, M. Neerincx, Evaluating XAI: A comparison of rule- based and example-based explanations, Artificial Intelligence (2021). doi: 10.1016/j.artint. 2020.103404

    2021

  22. [30]

    Colin, T

    J. Colin, T. FEL, R. Cadene, T. Serre, What i cannot predict, i do not understand: A human-centered evaluation framework for explainability methods, in: Advances in Neural Information Processing Systems, 2022. doi:10.48550/arXiv.2112.04417

    2022 doi arXiv

  23. [31]

    Fischbacher, z-Tree: Zurich toolbox for ready-made economic experiments, Experimental Economics (2007)

    U. Fischbacher, z-Tree: Zurich toolbox for ready-made economic experiments, Experimental Economics (2007). doi:10.1007/s10683-006-9159-4

    2007 doi

  24. [32]

    D. L. Chen, M. Schonger, C. Wickens, oTree—An open-source platform for laboratory, online, and field experiments, Journal of Behavioral and Experimental Finance (2016). doi: 10.1016/j.jbef. 2015.12.001

    2016 doi

  25. [33]

    D. A. Quispe G, F. Rajabiyazdi, G. A. Jamieson, A Machine Learning-Based Micro-World Platform for Condition-Based Maintenance, in: 2020 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2020. doi:10.1109/SMC42975.2020.9283448

    2020

  26. [34]

    Gentile, B

    D. Gentile, B. Donmez, G. A. Jamieson, Human performance consequences of normative and contrastive explanations: An experiment in machine learning for reliability maintenance, Artificial Intelligence (2023). doi:10.1016/j.artint.2023.103945

    2023

  27. [35]

    Vasconcelos, M

    H. Vasconcelos, M. Jörke, M. Grunde-McLaughlin, T. Gerstenberg, M. S. Bernstein, R. Krishna, Explanations Can Reduce Overreliance on AI Systems During Decision-Making, Proc. ACM Hum.-Comput. Interact. (2023). doi:10.48550/arXiv.2212.06823

    2023 doi arXiv

  28. [36]

    Buçinca, M

    Z. Buçinca, M. B. Malaya, K. Z. Gajos, To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-assisted Decision-making, Proceedings of the ACM on Human- Computer Interaction (2021). doi:10.1145/3449287

    2021 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.