The widespread use of Artificial Intelligence (AI) in various domains has led to a growing demand for algorithmic understanding, transparency, and trustworthiness. The field of eXplainable AI (XAI) aims to develop techniques that can inspect and explain AI systems' behaviour in a way that is understandable to humans. However, the efectiveness of explanations depends on how users perceive them, and their acceptability is connected with the level of understanding and compatibility with users' existing knowledge. So far, researchers in XAI have primarily focused on technical aspects of explanations, mostly without considering users' needs, and this aspect is necessary to consider for a trustworthy AI. In the meantime, there is a growing interest in human-centered approaches that focus on the intersection between AI and human-computer interaction, what is termed as human-centered XAI (HC-XAI). HC-XAI explores methods to achieve user satisfaction, trust, and acceptance for XAI systems. This paper presents a systematic survey on HC-XAI, reviewing 75 papers from various digital libraries. The contributions of this paper include: (1) identifying common human-centered approaches, (2) providing readers with insights into design perspectives of HC-XAI approaches, and (3) categorising with quantitative and qualitative analysis of all the papers under study. The findings stimulate discussions and shed light on ongoing and upcoming research in HC-XAI.
Over the last two decades, Artificial Intelligence (AI) has received an overwhelming response
from daily life applications, and industries such as autonomous vehicles, financial services, and
healthcare [
Notice that, the development of XAI applications is complex because the efectiveness of an
explanation depends on how the user perceives it [
In recent years, the XAI community has shown a growing interest in human-centered
approaches [12, 13, 14, 15, 16]. To address the crucial requirements regarding transparency,
understanding and trustworthiness in XAI systems, a new area of research has emerged where
AI and human-computer interaction (HCI) intersect [17]. This research field is termed
humancentered XAI (HC-XAI) [18]. It is important to examine the existing theoretical and practical
frameworks for generation and evaluation of explanations, regarding both efectiveness and
impact of explanations on user’s trust and satisfaction. The consideration for user requirements
and contextual factors in the design and evaluation of explanations are deemed necessary [
In this paper, we conducted a systematic survey on HC-XAI. In contrast to previous surveys, our approach paid major attention to research papers dealing with human-centered factors, including artefacts and techniques which strive to address the human needs for explanations. The survey focuses on reviewing 75 papers from Scopus, Web of Science, IEEE, and ACM digital libraries. Our objectives are: 1) to lay out common approaches for HC-XAI, 2) to provide insights on the design perspectives of common approaches, and 3) the categorisation of included papers. To achieve these objectives, we conducted a systematic literature review driven by the following specific research questions: • RQ1: What are the most common human-centered approaches (including frameworks, methods, theories, and artefacts) in XAI? • RQ2: How is the co-design perspective considered in HC-XAI approaches? • RQ3: How grounded are human-centered approaches on theoretical and practical aspects of explanations? Accordingly, we first categorise and provide insights into common approaches for HC-XAI. Then, we do quantitative and qualitative analyses of the papers under study. Our overarching goal is to bring attention to the ongoing and upcoming research related to HC-XAI. Therefore, the rest of the paper is organised as follows. Section 2 introduces background and related work. Section 3 details the methods used in the systematic literature review. Section 4 presents the most outstanding quantitative results. Section 5 provides a qualitative analysis of the most relevant papers. Section 6 discusses implications and research opportunities. Section 7 provides readers with final remarks.
Human-centered explanations are designed and tailored to meet the needs, understanding,
and cognitive abilities of human recipients [
On the one hand, there are numerous reviews available on the expanding field of XAI. They
play a vital role in defining and establishing the concept of XAI [
On the other hand, in recent years, there is a growing interest to highlight the needs and
importance of human-centered approaches [12, 13, 14, 15, 16]. To address this aspect, Chromik
and Schuessler [
In the following sections we will provide readers with a systematic literature survey to categorise existing approaches which introduce, devise, or apply human-centered artefacts and techniques for XAI.
In this section, we outline our methodology for identification, screening, eligibility (reviewing),
and inclusion of papers. To ensure comprehensive insights into the HC-XAI domain, we
conducted a systematic collection of papers, regarding both quantitative and qualitative information.
Our methodology follows the guidelines established by Moher et al. [
Publication Search Engine. In order to ensure a comprehensive and manageable selection of papers, we selected interdisciplinary databases (search engines) such as Web of Science (WoS), Scopus, ACM Digital Library, and IEEE digital Explorer for the initial selection of papers. These databases not only encompass research publications in the field of Computer Science, but also index studies in scientific fields such as AI, HCI, Social Sciences and Philosophy; enabling a comprehensive review of the literature related to our research questions.
Publication Year. We limited our selection to papers published between 2018 and mid-May, 2023. This time frame was chosen due to the significant surge of interest in XAI in recent years, particularly following the implementation of regulations such as the GDPR and AI act in Europe, as well as other AI-related acts in the UK and the US.
Search Strategy. On May 15th, 2023, we conducted the advanced search using web tools provided by the selected digital databases. These tools allow for the replication of the study and ensure consistent querying across all the databases. We performed a query consisting of multiple terms on the title, abstract, and author keywords.
The query for the digital databases was as follows: ((expla* AI OR expla* Artificial Intelligence OR Counterfactual expla*) OR (interpret* AI OR interpret* Artificial Intelligence)) AND (generat* OR framework* OR develop* OR (human-centered OR human-centred) OR (user-centered OR user-centred)). To maximise the diversity of the retrieved papers, we used word-stems in our search query. For example, the search item “expla*” covered all word-forms such as “explanation”, “explaining”, “explanatory”, etc. We used the search terms to gather the most recent publications that mention explainable AI and interpretable AI, respectively, across all subject areas. Also, the search query was overlapping to diferentiate between publications covering human-centered and user-centered development frameworks (to cover the broader perspective of human-centricity). The terms “generate”, “framework”, and “develop”, along with their corresponding word-forms, were used to appropriately limit the pool of publications in the unified set.
Inclusion and Exclusion. We focused on papers within the domain of HC-XAI. To be included, papers needed to present original work that introduced, applied, proposed, or evaluated human-centered methods or theories for explaining AI techniques, models, or systems. The initial filtering of papers (screening) was based on title, abstract and author keywords. Our paper selection process focused on original research articles that underwent peer-review and were written in English. Prior to evaluating the main content of each paper based on our inclusion criteria (eligibility), we manually excluded papers that comprised extended abstracts, doctoral consortium submissions, early career tracks, invited talks or tutorials, as well as all records which were not written in English.
PRISMA Guidelines. The flow diagram for the systematic literature review is shown in Figure 1, in which we illustrate diferent phases in agreement with the PRISMA guidelines.
Identification
Screening
Eligibility
Included Initial Search
Scopus (n=189)
WoS (n=41) ACM Digital Library (n=79) IEEE Digital Library (n=18)
Title and abstract screening (n=238)
Full text screening (n=82) Duplicates removed (n=89)
Papers excluded (n=156)
Papers excluded (n=19)
Papers covered in survey
n=(75) Papers included upon snowballing (n=12)
In the identification phase, we conducted an initial search across the selected digital libraries, using the predefined search query mentioned earlier. The results from this search yielded 18 papers from the IEEE digital library, 79 papers from the ACM digital library, 41 papers from WoS, and 189 papers from Scopus making a total of 327 papers. These relatively low numbers can be attributed to the limited research activity on HC-XAI.
In the screening phase, the relevant information (i.e., title, abstract, keywords, authors’ names, journal name, source, publisher, and year of publication) from the identified records was exported to a Microsoft Excel spreadsheet. A total of 238 records were left after 89 duplicates were removed. These 238 papers were then subjected to screening based on the inclusion criterion. After the screening process, 156 papers were deemed ineligible and excluded from further consideration. The remaining 82 papers entered the eligibility phase, where each paper underwent a thorough analysis through full-text reading. In the eligibility phase, a decision was made on each paper, determining whether it would proceed to the final phase. Unfortunately, 19 papers did not adequately address to our research questions and were excluded, leaving a total of 63 papers for the final phase.
During the final inclusion phase, 12 additional papers were included through a snowballing
process [
Review Process. We conducted a thorough analysis of the final set of papers, following the
taxonomy proposed by Guidotti et al. [
Before addressing the research questions and related answers, we conducted a bibliometric analysis on the aggregated query results. This analysis allowed us to gain a comprehensive understanding of the HC-XAI research field.
Figure 3 illustrates the growth of papers in the field of HC-XAI over the past years (from 2018
to mid-May 2023). The small number of papers published in 2018 (5) and 2019 (6), could be
attributed to the relatively lower popularity of HC-XAI during those years (which coincide
with the initial years of application of the GDPR) or to the limited usage of the related terms
(i.e., “human-centered” and “user-centered”) at that time. Anyway, there has been a consistent
increase in the number of papers since 2020, with a very significant rise in 2022, even if we do
not observe in HC-XAI the same exponential growth noticed by Adadi and Berrada [
The categorisation process considers various dimensions such as the type of data used, the specific task being addressed, the approach taken in contributing to the field (whether it involves application, evaluation, or development), and the specific type of explanation employed to clarify the underlying predictions or decisions made by the intelligent system.
Diferent data types are taken as input for predictive models such as image, text, and tabular
data. Our findings indicate that image data (8%) predominantly constitute the majority in
heatmap explanations, tabular data (57%) for feature scores and counterfactual explanations
while text input (6%) is primarily employed for textual explanations. The category “Other” (28%)
is introduced to account for such data types as time series, graphs, or logical rules (ontology) that
do not fit the predefined categories. Of course, it is important to note that these statistics may
be influenced by our choice of publication venues, and the complexity of the task itself can also
have an impact. Table 1 presents a quantitative analysis of various types of tasks, along with the
related data types for those tasks, in the included papers and their corresponding bibliographical
references. These are 49 papers out of 75 which solve any task in their contributing work.
As shown in Table 1, a significant portion (51%) of the papers focus on classification models.
This can be attributed to the broad nature of classification tasks and the popularity of posthoc
explanations, as explaining specific classification decisions serves as a compelling use case for
outcome explanations. We observed that classification tasks were performed by using mainly
tabular data, image or text, but also other data such as time series or graphs [
The second largest category of papers (29%) addressed both classification and regression
tasks concurrently in their studies. Notice that, papers which only deal with regression tasks
consider only tabular data [55, 56, 57]. In addition, papers which account for recommendation
tasks, they work with tabular [
On the other hand, we found out a diversity in the types of explanations under study. We categorised specific explanations in five main distinct categories which apply, devise, and evaluate (49 out of 75 papers), as outlined in Table 2. Notably, the most prevalent explanation type is the so-called hybrid approach which combines feature scores and counterfactual methods, which was found in 37% of the papers. This is followed by the use of counterfactuals alone (27%), feature scores alone (20%), and text highlighting (8%). Other explanation types include saliency maps, heatmaps, ontology, and rules, each accounting for 2% of the papers.
Moreover, the analysis also highlights the dominance of approaches for explaining black-box
models in current research on human-AI interaction. Local feature explanations, exemplified by
LIME [9] and SHAP [10], are commonly employed to answer the “why” question. Counterfactual
explanations, on the other hand, are used to tackle “what-if” or “why-not” questions. In the
context of recommendation systems, a hybrid approach combining text highlighting, rules,
and counterfactuals is also utilised, such as in a hybrid recommendation system [62]. Other
explanation types include text highlighting, saliency maps, heatmaps, rules, and an ontological
representation of explanations. Text highlighting, as referenced in papers [
The objective of this section is to provide insights to researchers and practitioners seeking appropriate contributions in the field of HC-XAI. In Table 3, we present the categorisation of papers that devise (introduce), apply, and evaluate the existing methods along with their specific contribution. It is worth noting that when a method devises an HC-XAI approach, it can be linked to multiple types of explanations, as there is no precondition of one-to-one correspondence between a method and an explanation type in the set of papers under study.
By examining the included papers, a categorisation of main contributions is presented in
Table 3. The first group of papers provides theoretical guidelines for designing HC-XAI
frameworks. The second group consists of user studies and interviews. We also refer to previous
user studies conducted on ML systems [
In this section, we go in depth with answering, from a qualitative viewpoint, the three research questions that we posed in the introduction.
RQ1 deals with frameworks, methods, theories and artefacts that support HC-XAI. First of all, we paid attention to generic concerns in the context of XAI (see Table 4).
The primary studies highlighted multiple objectives for XAI, with many papers emphasising the general reasons and prerequisites for their implementation. Fostering user trust and enhancing transparency emerged as the most prominent goals among the commonly identiifed objectives. Another recurring theme observed in the studies revolved around catering to the user satisfaction as well as user needs, user experience, and social needs across diverse domains. This emphasis stems from the fact that diferent users (including domain experts, end-consumers, AI engineers, legislators...) possess unique motivations for seeking explanations. The purpose of XAI is to efectively assist each user in achieving their specific objectives by tailoring the explanations to their individual needs. User-centered approaches ensure that XAI accommodates properly diverse requirements and aspirations of users across various domains. We only found a few papers to aid user future actions and promoting informed decision-making. Other recurring objectives which require further attention are related to the improvement of the human-AI collaboration and thus increasing the task performance.
RQ2 investigated XAI techniques that are supported by a co-design perspective, i.e., they incorporate interaction or dialogue as a human-centered element and play a crucial role in enhancing transparency and understanding of AI systems. These techniques aim to bridge the gap between complex ML models and human users by enabling meaningful conversations and explanations. Through the integration of dialogue, XAI techniques foster a more inclusive and user-centered approach, ensuring that AI systems are not black boxes but rather tools that empower individuals to make informed decisions. For example, by allowing users to engage in a dialogue with an AI system, Stepin et al. [98] facilitate a two-way exchange of information through an information-seeking dialogue, where the AI system provides explanations for its decisions and users can seek clarification or express their concerns.
In addition, Akula et al. [
RQ3 investigated HC-XAI methods which are grounded on theoretical and practical evaluation approaches. Evaluation of explanations can be supported by functionally-grounded, humangrounded, or application-grounded approaches (see Table 5). We adhere to the categorisation of metrics proposed by Doshi-Velez and Kim [19].
Functionally-grounded metrics do not require any human feedback. They are based on objective criteria that can be measured without human involvement. As a result, they measure the formal properties of the explainer, while regarding functional aspects of explanations (e.g., ifdelity, accuracy, actionability, sparsity, or plausibility). Human-grounded metrics require direct human involvement for their measurement. They emphasise the human perspective and consider cognitive and psychological factors (e.g., satisfaction, persuasiveness, or novelty). These metrics gather feedback from users to assess the quality and usefulness of explanations. Applicationgrounded metrics focus on real-world applicability and impact. Their computation usually involves domain experts who are asked to assess how well the intelligent system performs within specific application domains, considering practical constraints and requirements.
Insights from the systematic literature review carried out in the previous sections indicate a pressing need to develop human-centered explanation approaches in various fields of AI. Traditional methods of explaining AI decisions often fall short in meeting human requirements for transparency and understanding. However, a promising avenue that has gained considerable attention is the utilisation of human-perspective for explanation generation and evaluation.
Human-centered explanations play a critical role in promoting transparency, trust, and accountability in AI systems. Such explanations enable users to understand the reasoning behind automated decisions and empower them to make informed choices. However, existing research indicates a gap in actionable and human-centered explanations, what leads to some research opportunities for the exploration of human-centered approaches in the field of HC-XAI: • Bridging the explanation gap by handling user needs. The systematic literature review reveals a persistent gap in providing actionable explanations that are meaningful and relevant to users. The explanations driven by user feedback ofer a promising approach to bridge the gap. By incorporating user preferences and context, explanations can provide actionable insights that resonate with users and facilitate more informed decision-making. • Personalisation and Contextualisation. Diferent users may have diverse preferences, values, and needs, and a one-size-fits-all explanation may not be suficient. By involving users in the explanation generation process, explanations can adapt to individual user characteristics and provide insights that are tailored to their specific context. • Closing the Feedback Loop. The literature review indicates a need to close the feedback loop between users and AI systems, enabling iterative improvements and accountability. This iterative feedback loop can enhance system performance, address biases and limitations, but also ensure the system evolves based on user needs and expectations. • Human-AI Collaboration and Co-creation. Human-centred explanations promote human-AI collaboration and co-creation. By involving users in the generation and evaluation of explanations, we facilitate a meaningful dialogue between users and AI systems. Users can: provide feedback, ask questions, actively participate in refining and improving the system’s behaviour.
In this work, we have presented a systematic literature review from which we can conclude that there is a need for more human-centered explanation approaches. Such HC-XAI approaches can enhance human understanding, enable ethical evaluations, and foster user engagement, thereby serving the “human-in-the-loop” cause. Moreover, building human-centered explanations means bridging the explanation gap, enhancing user agency, enabling personalisation and contextualisation, addressing trust and ethical concerns, closing the feedback loop, and promoting human-AI collaboration. Consequently, we can bridge the gap between complex algorithms and human intuition, empowering individuals to make informed decisions, identifying biases, and actively taking part in the development of responsible AI technologies.
Even though we conducted a medium-scale survey, it was more feasible and practical than larger surveys, providing valuable insights into human-centered approaches, while requiring fewer resources, such as time and human resources. Despite its limitations, this survey can provide a starting point for further research and help determine the viability of pursuing largerscale surveys. In conclusion, various factors can influence the efectiveness of explanations in human-centered perspectives. These factors include the specific explanation technique used, the characteristics of the dataset, and the nature of the task at hand.
Muhammad Sufian is a PhD researcher (Matricola N.309445). Ilia Stepin is an FPI researcher (grant number: PRE2019-090153). This research was funded by MCIN/AEI/10.13039/501100011033 (grants PID2021-123152OB-C21, TED2021-130295B-C33, and RED2022-134315-T), the Galician Ministry of Culture, Education, Professional Training, and University (grants ED431C2022/19 and ED431G2019/04). All grants were co-funded by the European Regional Development Fund (ERDF/FEDER program). [7] C. Rudin, Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead, Nature machine intelligence 1 (2019) 206–215. [8] X. He, Y. Hong, X. Zheng, Y. Zhang, What are the users’ needs? design of a user-centered explainable artificial intelligence diagnostic system, International Journal of Human– Computer Interaction 39 (2023) 1519–1542. [9] M. T. Ribeiro, S. Singh, C. Guestrin, Model-agnostic interpretability of machine learning, arXiv preprint arXiv:1606.05386 (2016). [10] S. M. Lundberg, S.-I. Lee, A unified approach to interpreting model predictions, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, Curran Associates Inc., Red Hook, NY, USA, 2017, p. 4768–4777. [11] M. Ribera, A. Lapedriza, Can we do better explanations? a proposal of user-centered explainable AI, in: IUI workshops, volume 2327, 2019, p. 38. [12] U. Ehsan, M. O. Riedl, Human-centered explainable AI: Towards a reflective sociotechnical approach, in: HCI International 2020-Late Breaking Papers: Multimodality and Intelligence: 22nd HCI International Conference, HCII 2020, Copenhagen, Denmark, July 19–24, 2020, Proceedings 22, Springer, 2020, pp. 449–466. [13] H. Chen, C. Gomez, C.-M. Huang, M. Unberath, Explainable medical imaging AI needs human-centered design: guidelines and evidence from a systematic review, npj Digital Medicine 5 (2022) 156. [14] J. Graefe, S. Paden, D. Engelhardt, K. Bengler, Human centered explainability for intelligent vehicles–a user study, in: Proceedings of the 14th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, 2022, pp. 297–306. [15] T. Miller, Explanation in artificial intelligence: Insights from the social sciences, Artificial intelligence 267 (2019) 1–38. [16] M. Sufian, P. Graziani, J. M. Alonso, A. Bogliolo, FCE: Feedback based counterfactual explanations for explainable AI, IEEE Access 10 (2022) 72363–72372. [17] Q. V. Liao, M. Pribić, J. Han, S. Miller, D. Sow, Question-driven design process for explainable
AI user experiences, arXiv preprint arXiv:2104.03483 (2021). [18] U. Ehsan, P. Wintersberger, Q. V. Liao, M. Mara, M. Streit, S. Wachter, A. Riener, M. O.
Riedl, Operationalizing human-centered perspectives in explainable AI, in: Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, 2021, pp. 1–6. [19] F. Doshi-Velez, B. Kim, Towards a rigorous science of interpretable machine learning, arXiv preprint arXiv:1702.08608 (2017). [20] S. Mohseni, N. Zarei, E. D. Ragan, A multidisciplinary survey and framework for design and evaluation of explainable AI systems, ACM Transactions on Interactive Intelligent Systems (TiiS) 11 (2021) 1–45. [21] J. J. Ferreira, M. S. Monteiro, What are people doing about XAI user experience? a survey on AI explainability research and practice, in: Design, User Experience, and Usability. Design for Contemporary Interactive Environments: 9th International Conference, DUXU 2020, Held as Part of the 22nd HCI International Conference, HCII 2020, Copenhagen, Denmark, July 19–24, 2020, Proceedings, Part II 22, Springer, 2020, pp. 56–73. [22] D. Wang, Q. Yang, A. Abdul, B. Y. Lim, Designing theory-driven user-centric explainable AI, in: Proceedings of the 2019 CHI conference on human factors in computing systems, do not behave like users expect, in: Proceedings of the 30th ACM Conference on User Modeling, Adaptation and Personalization, 2022, pp. 110–120. [55] H. Ma, K. McAreavey, R. McConville, W. Liu, Explainable AI for non-experts: Energy tarif forecasting, in: 2022 27th International Conference on Automation and Computing (ICAC), IEEE, 2022, pp. 1–6. [56] C. Bove, J. Aigrain, M.-J. Lesot, C. Tijus, M. Detyniecki, Contextualization and exploration of local feature importance explanations to improve understanding and satisfaction of non-expert users, in: 27th international conference on intelligent user interfaces, 2022, pp. 807–819. [57] B. Hayes, M. Moniz, Trustworthy human-centered automation through explainable AI and high-fidelity simulation, in: Advances in Simulation and Digital Human Modeling: Proceedings of the AHFE 2020 Virtual Conferences on Human Factors and Simulation, and Digital Human Modeling and Applied Optimization, July 16-20, 2020, USA, Springer, 2021, pp. 3–9. [58] M. Förster, P. Hühn, M. Klier, K. Kluge, User-centric explainable AI: design and evaluation of an approach to generate coherent counterfactual explanations for structured data, Journal of Decision Systems (2022) 1–32. [59] D. Cirqueira, M. Helfert, M. Bezbradica, Towards design principles for user-centric explainable AI in fraud detection, in: Artificial Intelligence in HCI: Second International Conference, AI-HCI 2021, Held as Part of the 23rd HCI International Conference, HCII 2021, Virtual Event, July 24–29, 2021, Proceedings, Springer, 2021, pp. 21–40. [60] A. A. Shrotri, N. Narodytska, A. Ignatiev, K. S. Meel, J. Marques-Silva, M. Y. Vardi, Constraint-driven explanations for black-box ML models, in: Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 2022, pp. 8304–8314. [61] H. Baniecki, D. Parzych, P. Biecek, The grammar of interactive explanatory model analysis,
Data Mining and Knowledge Discovery (2023) 1–37. [62] A. Silva, M. Schrum, E. Hedlund-Botti, N. Gopalan, M. Gombolay, Explainable artificial intelligence: Evaluating the objective and subjective impacts of XAI on human-agent interaction, International Journal of Human–Computer Interaction 39 (2023) 1390–1404. [63] Q. V. Liao, D. Gruen, S. Miller, Questioning the AI: informing design practices for explainable AI user experiences, in: Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, 2020, pp. 1–15. [64] J. E. Sales, A. Freitas, S. Handschuh, A user-centred analysis of explanations for a multicomponent semantic parser, in: Natural Language Processing and Information Systems: 25th International Conference on Applications of Natural Language to Information Systems, NLDB 2020, Saarbrücken, Germany, June 24–26, 2020, Proceedings 25, Springer, 2020, pp. 37–44. [65] F. Sovrano, F. Vitali, Explanatory artificial intelligence (YAI): human-centered explanations of explainable AI and complex data, Data Mining and Knowledge Discovery (2022) 1–28. [66] C. Panigutti, A. Beretta, D. Fadda, F. Giannotti, D. Pedreschi, A. Perotti, S. Rinzivillo, Codesign of human-centered, explainable AI for clinical decision support, ACM Transactions on Interactive Intelligent Systems (2023). [67] D. H. Kim, E. Hoque, M. Agrawala, Answering questions about charts and generating visual explanations, in: Proceedings of the 2020 CHI conference on human factors in (COMPASS), 2022, pp. 439–452. [96] A. Bertrand, R. Belloum, J. R. Eagan, W. Maxwell, How cognitive biases afect XAI-assisted decision-making: A systematic review, in: Proceedings of the 2022 AAAI/ACM conference on AI, ethics, and society, 2022, pp. 78–91. [97] M. Sufian, M. Y. Khan, A. Bogliolo, Towards human cognition level-based experiment design for counterfactual explanations, in: 2022 Mohammad Ali Jinnah University International Conference on Computing (MAJICC), 2022, pp. 1–5. doi:10.1109/MAJICC56935. 2022.9994203. [98] I. Stepin, K. Budzynska, A. Catala, M. Pereira-Fariña, J. M. Alonso-Moral, Informationseeking dialogue for explainable artificial intelligence: Modelling and analytics, Argument & Computation (2023).