Explainable Artificial Intelligence (XAI) from a user perspective: A synthesis of prior literature and problematizing avenues for future research

Haque, Akm Bahalul; Islam, A. K. M. Najmul; Mikalef, Patrick

doi:10.1016/j.techfore.2022.122120

Cited by 73 publications

(50 citation statements)

References 79 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This often has to do with the interpretability of the input data and how the explanation is conveyed (visuals, text, etc.) (Hudon et al, 2021;Haque et al, 2023).…”

Section: Human Interpretablementioning

confidence: 99%

“…As human-centric tasks often use tabular or time series data, their subsequent explanations are often not concise, actionable or interpretable easily beyond the scope of a data scientist's knowledge (Karran, Demazure, Hudon, Senecal, & Léger, 2022). Recent research on explanation user design has shown that humans across healthcare, law, finance, education, and e-commerce, among others, prefer hybrid text and visual explanations (Haque et al, 2023), a format not easily provided by current post-hoc libraries. Lastly, the consistency of the explanations is not intrinsically measured; generating an explanation for the next step in the time series could vary greatly from the previous step.…”

Section: Explainers Of Today: State-of-the-art and Limitationsmentioning

confidence: 99%

See 1 more Smart Citation

Trusting the Explainers: Teacher Validation of Explainable Artificial Intelligence for Course Design

Swamy

Marras

et al. 2023

LAK23: 13th International Learning Analytics and Knowledge Conference

View full text Add to dashboard Cite

Explainable Artificial Intelligence (XAI) plays a crucial role in enabling human understanding and trust in deep learning systems, often defined as determining which features are most important to a model's prediction. As models get larger, more ubiquitous, and pervasive in aspects of daily life, explainability is necessary to avoid or minimize adverse effects of model mistakes. Unfortunately, current approaches in human-centric XAI (e.g. predictive tasks in healthcare, education, or personalized ads) tend to rely on a single explainer. This is a particularly concerning trend when considering that recent work has identified systematic disagreement in explainability methods when applied to the same points and underlying black-box models. In this paper, we therefore present a call for action to address the limitations of current state-of-the-art explainers. We propose to shift from post-hoc explainability to designing interpretable neural network architectures; moving away from approximation techniques in human-centric and high impact applications. We identify five needs of human-centric XAI (real-time, accurate, actionable, human-interpretable, and consistent) and propose two schemes for interpretable-by-design neural network workflows (adaptive routing for interpretable conditional computation and diagnostic benchmarks for iterative model learning). We postulate that the future of human-centric XAI is neither in explaining black-boxes nor in reverting to traditional, interpretable models, but in neural networks that are intrinsically interpretable.

show abstract

“…This often has to do with the interpretability of the input data and how the explanation is conveyed (visuals, text, etc.) (Hudon et al, 2021;Haque et al, 2023).…”

Section: Human Interpretablementioning

confidence: 99%

Section: Explainers Of Today: State-of-the-art and Limitationsmentioning

confidence: 99%

Trusting the Explainers: Teacher Validation of Explainable Artificial Intelligence for Course Design

Swamy

Marras

et al. 2023

LAK23: 13th International Learning Analytics and Knowledge Conference

View full text Add to dashboard Cite

show abstract

“…Conversational AI agents, such as Replika and Alexa, have become context-aware such that they can adapt and engage in prolonged conversations and understand users’ emotions (Chaturvedi et al , 2023; Skjuve et al , 2021). Moreover, the current wave of advancements shifted from affective computing to generative AI, leading to the emergence of complex agents such as ChatGPT, whose self-supervised learning algorithms can create new content and discuss any topic (Dwivedi et al , 2023; Chaturvedi et al , 2023). Such agents can assume different roles, such as customer service agents, and have already caused digital disruption in the marketplace.…”

Section: Conceptual Backgroundmentioning

confidence: 99%

“…Although our data primarily focused on English-speaking countries, existing research on other conversational AI agents, such as Xiaoce in Asian countries, supports our findings. For instance, Chaturvedi et al (2023) and Shum et al (2018) discuss how Chinese and Japanese consumers rely on the conversational AI agent Xiaoice to provide emotional support and alleviate loneliness, while anecdotal evidence suggests that heavy users of this agent spend longer time interacting with it than with other fellow humans (Gaubert, 2021). Nonetheless, we recommend that further research examine how consumers from different cultures form relationships with conversational AI agents and what the role of self-related processes is in such scenarios.…”

Section: Limitations and Future Researchmentioning

confidence: 99%

More than just a chat: a taxonomy of consumers’ relationships with conversational AI agents and their well-being implications

Alabed,

Javornik,

Gregory-Smith

et al. 2023

EJM

View full text Add to dashboard Cite

Purpose This paper aims to study the role of self-concept in consumer relationships with anthropomorphised conversational artificially intelligent (AI) agents. First, the authors investigate how the self-congruence between consumer self-concept and AI and the integration of the conversational AI agent into consumer self-concept might influence such relationships. Second, the authors examine whether these links with self-concept have implications for mental well-being. Design/methodology/approach This study conducted in-depth interviews with 20 consumers who regularly use popular conversational AI agents for functional or emotional tasks. Based on a thematic analysis and an ideal-type analysis, this study derived a taxonomy of consumer–AI relationships, with self-congruence and self–AI integration as the two axes. Findings The findings unveil four different relationships that consumers forge with their conversational AI agents, which differ in self-congruence and self–AI integration. Both dimensions are prominent in replacement and committed relationships, where consumers rely on conversational AI agents for companionship and emotional tasks such as personal growth or as a means for overcoming past traumas. These two relationships carry well-being risks in terms of changing expectations that consumers seek to fulfil in human-to-human relationships. Conversely, in the functional relationship, the conversational AI agents are viewed as an important part of one’s professional performance; however, consumers maintain a low sense of self-congruence and distinguish themselves from the agent, also because of the fear of losing their sense of uniqueness and autonomy. Consumers in aspiring relationships rely on their agents for companionship to remedy social exclusion and loneliness, but feel this is prevented because of the agents’ technical limitations. Research limitations/implications Although this study provides insights into the dynamics of consumer relationships with conversational AI agents, it comes with limitations. The sample of this study included users of conversational AI agents such as Siri, Google Assistant and Replika. However, future studies should also investigate other agents, such as ChatGPT. Moreover, the self-related processes studied here could be compared across public and private contexts. There is also a need to examine such complex relationships with longitudinal studies. Moreover, future research should explore how consumers’ self-concept could be negatively affected if the support provided by AI is withdrawn. Finally, this study reveals that in some cases, consumers are changing their expectations related to human-to-human relationships based on their interactions with conversational AI agents. Practical implications This study enables practitioners to identify specific anthropomorphic cues that can support the development of different types of consumer–AI relationships and to consider their consequences across a range of well-being aspects. Originality/value This research equips marketing scholars with a novel understanding of the role of self-concept in the relationships that consumers forge with popular conversational AI agents and the associated well-being implications.

show abstract

“…Despite their transformative potential, the inner mechanics of AI remain obscured. Such systems often resemble a “black box”, presenting limited interpretability regarding decision-making processes. , …”

Section: Introductionmentioning

confidence: 99%

Optimizing the UV-Fenton Degradation of m-Cresol Wastewater: An Experimental and Artificial Intelligence Modeling Approach

Zhang,

Yao,

Zhao

et al. 2024

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Wastewater treatment, especially the efficient degradation of contaminants such as m-cresol, remains a pivotal challenge. This study investigates the application of artificial neural networks (ANN) in predicting total organic carbon (TOC) removal rates from m-cresol-contaminated wastewater by using the ultraviolet (UV)-Fenton oxidation process. Six key variables, namely, Fe 2+ dosage, H 2 O 2 dosage, catalyst quantity, reaction time, pH, and substrate concentration, were employed as inputs to the ANN model. Leveraging this multivariable input and a comprehensive data set, the ANN model projected a maximum TOC removal rate of 87.12%, validated by an efficiency of 86.26% achieved through experiments under the derived optimal conditions: Fe 2+ dosage at 16.09 mg/L, H 2 O 2 dosage at 1.40 mg/L, catalyst quantity at 0.11 g/L, reaction time of 29.80 min, initial pH of 3.66, and substrate concentration of 50 mg/L. Comparative analysis with other machine learning algorithms further revealed that the ANN model notably outperformed linear regression, support vector regression, and random forest in terms of precision. This work paves the way for resource-optimized experimental designs, fostering real-time wastewater monitoring and refining advanced oxidation process proficiency in industrial applications.

show abstract

Explainable Artificial Intelligence (XAI) from a user perspective: A synthesis of prior literature and problematizing avenues for future research

Cited by 73 publications

References 79 publications

Trusting the Explainers: Teacher Validation of Explainable Artificial Intelligence for Course Design

Trusting the Explainers: Teacher Validation of Explainable Artificial Intelligence for Course Design

More than just a chat: a taxonomy of consumers’ relationships with conversational AI agents and their well-being implications

Optimizing the UV-Fenton Degradation of m-Cresol Wastewater: An Experimental and Artificial Intelligence Modeling Approach

Contact Info

Product

Resources

About