Textual Explanations Research Articles

The Explanation That Hits Home: The Characteristics of Verbal Explanations That Affect Human Perception in Subjective Decision-Making

Human-AI collaborative decision-making can achieve better outcomes than either party individually. The success of this collaboration can depend on whether the human decision-maker perceives the AI contribution as beneficial to the decision-making process. Beneficial AI explanations are often described as relevant, convincing, and trustworthy. Yet, we know little about the characteristics of explanations that result in these perceptions. Focusing on collaborative subjective decision-making, using the context of subtle sexism, where explanations can surface new interpretations, we conducted a user study (N=20) to explore the structural and content characteristics that affect perceptions of human and AI-generated verbal (text and audio) explanations. We find four groups of characteristics ( Tone, Grammatical Elements, Argumentative Sophistication and Relation to User ), and that the effect of these characteristics on the perception of explanations for subtle sexism depends on the perceived author. Thus, we also identify which explanation characteristics participants use to identify the author of an explanation. Demonstrating the relationship between these characteristics and explanation perceptions, we present a categorized set of characteristics that system builders can leverage to produce the appropriate perception of an explanation for various sensitive contexts. We also highlight human perception biases and associated issues resulting from these perceptions.

Education of the "SusCatinTing" application on changes in knowledge, attitudes, and behavior of brides-to-be

The "SusCatinTing" application is a digital health intervention designed to enhance the knowledge, attitudes, and practices (KAP) of brides-to-be regarding stunting prevention during the critical first 1000 days of life. This study aimed to evaluate the effectiveness of this application in improving stunting prevention efforts among brides-to-be in Bengkulu Province, Indonesia. The application provides educational content on healthy pregnancy, maternal nutrition, and stunting prevention strategies, incorporating both text and audio explanations to ensure accessibility for illiterate and visually impaired users. A quantitative research design was employed, utilizing a single-group pretest-posttest approach involving 50 pairs of brides-to-be with internet-connected Android smartphones. Data collection was conducted through structured questionnaires administered by trained enumerators, focusing on KAP related to stunting prevention. The results indicated a significant improvement in KAP scores post-intervention (knowledge score increase: 42.59, p < 0.001). The application’s unique features, including interactive quizzes and a certification process for marriage registration at the Office of Religious Affairs, contributed to its effectiveness. The study concludes that the "SusCatinTing" application is a promising tool for enhancing health education and promoting preventive behaviors among brides-to-be. Further research is recommended to explore the long-term impact of such digital interventions on stunting rates and maternal-child health outcomes.

Gamification of virtual museum curation: a case study of Chinese bronze wares

Museums, which are among the most popular science institutions outside schools, are usually used to display and introduce historical culture and cultural relics to tourists. Text and audio explanations are used by traditional museums to popularize historical knowledge and science for tourists, and general interactive systems are based on desktops. This learning method is relatively boring in terms of experience. As a result, tourists have no desire or interest in actively exploring and learning about bronze ware, so they only have a basic understanding about bronze ware. Since most tourists are familiar with games, they are more likely to be attracted by game content and will actively explore and interact with it. In addition, a certain degree of reality is created by virtual reality technology and an immersive experience through head-mounted devices is provided to users. In this paper, we take Chinese bronzes as the research objects. We first use 3D laser scanners to obtain bronze models ; then, we build a virtual museum environment, and we finally design a virtual reality curation game based on this bronze digital museum. This game offers visitors an immersive museum roaming and bronze ware interactive experience. Through a combination of text, video learning, and games, visitors’ curiosity and desire to explore bronze ware are stimulated, and their understanding and ability to remember bronze ware knowledge can be deepened. In terms of cultural heritage, this game is also conducive to the spread of traditional Chinese bronze culture throughout the world.

Can You Tell Real from Fake Face Images? Perception of Computer-Generated Faces by Humans

With recent advances in machine learning and big data, it is now possible to create synthetic images that look real. Face generation is often of particular interest, as faces can be used for various purposes. However, improper use of such content can lead to the dissemination of false information, such as fake news, and thus pose a threat to society. This work studies whether people believe the truthfulness of faces using eye tracking and self-reports, including free-form textual explanations when participants encounter real and computer-generated faces. We used three different datasets for our evaluations, and our experimental results show that while people are relatively better at identifying the truthfulness of real faces and faces generated by earlier machine learning algorithms with different gazing behaviors in viewing and rating phases, they perform less accurately when deciding the truthfulness of synthetic face images that are generated by newer algorithms. Our findings provide important insights for society and policymakers.

Visualizing thyroid health: a pictorial journey through 2017 ACR TI-RADS and common thyroid pathologies.

With the advent of high-resolution ultrasonography (HRUS), more thyroid nodules are being detected than ever before, and they are being identified at an earlier stage. It poses a challenge for radiologists and clinicians in deciding what to do next. Most nodules are benign and require no follow-up and intervention. Even highly suspicious nodules can be followed up, if the size is small. Variations in HRUS interpretation among radiologists are common, with frequent misidentifications between spongiform and solid-cystic lesions, hypoechoic and very hypoechoic nodules, and microcalcification and hyperechoic foci with comet-tail artifacts. Cystic lesions with echogenic contents are often confused with solid nodules, cystic papillary carcinoma thyroid is often confused with colloid cysts. The 2017 ACR TI-RADS (American College of Radiology Thyroid Imaging Reporting and Data System) aims to standardize the interpretation of thyroid nodules and guide further management. Rather than giving specific diagnosis like colloid cyst, adenomatous nodule and papillary carcinoma; ACR TI-RADS classifies nodules from TI-RADS 1 to TI-RADS 5 based on HRUS characteristics and recommends further management. What the authors often read are textual contents that are theoretical, and in practice, the authors get confused while interpreting the characteristics of thyroid nodules. This review offers a detailed visual overview of the 2017 ACR TI-RADS and common thyroid conditions, explaining key features through imaging data and examples for consistent interpretation. Combining textual explanations with visual aids, this article provides practical guidance for interpreting thyroid nodules for radiologists, and clinicians seeking a clear understanding of thyroid imaging and pathology.

Investigating multimodal scientific communication: an analysis of communicative modes in Biology and Engineering research articles

The dissemination of knowledge in science and technology relies on efficient communication, with a noticeable shift towards multimodal communication. This research explores the subtle patterns of multimodal scientific communication to comprehend how several modes work together to enhance discourse quality. Twenty research articles, ten from the field of Engineering and ten from Biology, written over the previous three years by researchers of the University of El Oued, Algeria were collected from credible journals and then examined using a qualitative content analysis. The findings demonstrated that the authors used a multimodal approach in their research articles, presenting their research outcomes through a combination of textual explanations and visual components including graphs, tables, images, and diagrams. The study’s findings highlighted the interdependent nature of text and images and their importance in communicating scientific discourse. Furthermore, the data displays differences in the modes selected by these researchers, which reflects the specificities and particular needs of each discipline in choosing the modes. The study suggests that researchers’ active participation in workshops and training sessions can improve their proficiency in multimodal scientific communication. Moreover, it recommends additional research to broaden the area of enquiry and improve knowledge of multimodal communication in scientific discourse.

Co-Vision: Automated Video Context Understanding and Classification Using AI

In today's digital landscape, a challenge is presented by the ever-growing volume of video content: how to efficiently understand, access, and utilize this wealth of information. This challenge is addressed through developing an innovative web application combining video understanding, content recommendation, and accessibility features. The web application's core is found in its ability to be automatically analyzed and comprehended by videos' content. Valuable insights are extracted, key events, objects, and actions are identified, and textual explanations are generated, all of which are leveraged by state-of-the-art machine learning techniques in computer vision and natural language processing. The comprehension of video content forms the foundation for two significant applications, such as content recommendation, in which the content of videos and user preferences are understood, and our application revolutionizes content recommendation systems. Personalized video recommendations are provided, ensuring that content is discovered by users that precisely match their interests and needs. This user experience enhancement, along with resulting higher user engagement and content consumption on recommendation-driven platforms, is achieved. Accessibility is increased as the project promotes inclusivity by making video content more accessible to a broader audience. For individuals with disabilities, textual explanations and transcripts are generated by our application, breaking down barriers to understanding. Additionally, the convenience of summarizing lengthy videos is offered, enabling users to quickly grasp key insights without the need to watch the entire content. This enhanced accessibility is particularly valuable in educational contexts and beyond, where content consumption is made more efficient and equitable.

Undergraduate and graduate students' conceptual understanding of model classification outcomes under the lens of scientific argumentation

AbstractRecent advancements in artificial intelligence (AI) and machine learning (ML) have driven research and development across multiple industries to meet national economic and technological demands. Consequently, companies are investing in AI, ML, and data analytics workforce development efforts to digitalize operations and enhance global competitiveness. As such, evidence‐based educational research around ML is essential to provide a foundation for the future workforce as they face complex AI challenges. This study explored students' conceptual ML understanding through a scientific argumentation framework, where we examined how they used evidence and reasoning to support claims about their ML models. This framework lets us gain insight into students' conceptualizations and helped scaffold student learning via a cognitive apprenticeship model. Thirty students in a mechanical engineering classroom at Purdue University experimented with neural network ML models within a computational notebook to create visual claims (ML models) with textual explanations of their evidence and reasoning. Accordingly, we qualitatively analyzed their learning artifacts to examine their underfit, fit, and overfit models and explanations. It was found that some students tended toward technical explanations while others used visual explanations. Students with technically dominant explanations had higher proficiency in generating correctly fit models but lacked explanatory evidence. Conversely, students with visually dominant explanations provided evidence but lacked technical reasoning and were less accurate in identifying fit models. We discuss implications for both groups of students and offer future research directions to examine how positive pedagogical elements of learning design can optimize ML educational material and AI workforce development.

Information-seeking dialogue for explainable artificial intelligence: Modelling and analytics

Explainable artificial intelligence has become a vitally important research field aiming, among other tasks, to justify predictions made by intelligent classifiers automatically learned from data. Importantly, efficiency of automated explanations may be undermined if the end user does not have sufficient domain knowledge or lacks information about the data used for training. To address the issue of effective explanation communication, we propose a novel information-seeking explanatory dialogue game following the most recent requirements to automatically generated explanations. Further, we generalise our dialogue model in form of an explanatory dialogue grammar which makes it applicable to interpretable rule-based classifiers that are enhanced with the capability to provide textual explanations. Finally, we carry out an exploratory user study to validate the corresponding dialogue protocol and analyse the experimental results using insights from process mining and argument analytics. A high number of requests for alternative explanations testifies the need for ensuring diversity in the context of automated explanations.

QUALITY ENHANCEMENT OF THE STREET CORRIDOR TO SUPPORT TOURISM BASED ON URBAN DESIGN CONCEPT IN PEMUDA STREET, MAGELANG CITY

Pemuda Street is a corridor in the Magelang Chinatown area, known as one of the shopping centers in Magelang City. This area is generally a commerce and service area consisting of shops, healthcare facilities, and banking services. Despite its dominance by various economic opportunities and claims to be a Chinatown area, it requires a suitable area identity. Problems arise from various perspectives, such as pedestrian facilities being limited to pedestrians with unattractive visual appeal that fails to represent the distinctive characteristics of Chinatown. Consequently, motorcycle riders tend to utilize pedestrians more than walkers. Furthermore, most shops have experienced a decrease in visitors, influencing the economic earnings of the area. The Pemuda Street corridor aims to be planned using the design thinking method, adapting the concept of complete streets with elaboration from the concepts of commercial street, shared street, urban tourism, and vernacular settlement, to address various issues. The planning process begins with problem analysis and literature review to manifest the ideal planning concept. It concludes with creating a master plan design for the area summarized in textual explanations, two-dimensional images, and three-dimensional frameworks. The results of this research is apply the "complete street" planning concept can become the basis for designs that focus on functionality, aesthetics, sustainability, social needs, environmental harmony, and economic progress while respecting the historical values of the area Chinatown in Magelang City.

Improving Trust in AI with Mitigating Confirmation Bias: Effects of Explanation Type and Debiasing Strategy for Decision-Making with Explainable AI

With advancements in artificial intelligence (AI), explainable AI (XAI) has emerged as a promising tool for enhancing the explainability of complex machine learning models. However, the explanations generated by an XAI may lead to cognitive biases among human users. To address this problem, this study aims to investigate how to mitigate users’ cognitive biases based on their individual characteristics. In the literature review, we found two factors that can be helpful in remedying biases: 1) debiasing strategies that have been reported to potentially reduce biases in users’ decision-making via additional information or change in information delivery, and 2) explanation modality types. To examine these factors’ effects, we conducted an experiment with a 4 (debiasing strategy) × 3 (explanation type) between-subject design. In the experiment, participants were exposed to an explainable interface that provides an AI’s outcomes with explanatory information, and their behavioral and attitudinal responses were collected. Specifically, we statistically examined the effects of textual and visual explanations on users’ trust and confirmation bias toward AI systems, considering the moderating effects of debiasing methods and watching time. The results demonstrated that textual explanations lead to higher trust in XAI systems compared to visual explanations. Moreover, we found that textual explanations are particularly beneficial for quick decision-makers to evaluate the outputs of AI systems. Next, the results indicated that the cognitive bias can be effectively mitigated by providing users with a priori information. These findings have theoretical and practical implications for designing AI-based decision support systems that can generate more trustworthy and equitable explanations.

EGLR: Two-staged Explanation Generation and Language Reasoning framework for commonsense question answering

In the field of natural language processing, strong commonsense reasoning abilities are often required for intelligent systems to excel in the task of commonsense question answering (QA). To enhance the interpretability of QA systems, a natural approach is to generate textual explanations in addition to the predicted answers, ensuring the understandability of the results. Recent efforts address this by using a prompted language model (LM) with frozen parameters to generate explanations in a few-shot manner. These explanations are then used as additional context to guide a finetuned LM in making the final decision. However, these methods still underutilize the semantic information embedded in the explanatory text. Consequently, the reasoning models tend to rely on word co-occurrence and the knowledge stored in the model itself rather than fully exploiting the explanations. Therefore, we propose a two-staged Explanation Generation and Language Reasoning Framework (EGLR), Our framework takes advantage of the in-context learning capability of LMs to generate explanations and reformulates the reasoning task based on explanations as a semantic matching problem. Through joint prompting and training, our model can select the most appropriate explanation by comparing multiple explanations. Experimental results on three public datasets demonstrate that our framework achieves superior performance on the full dataset while maintaining performance on out-of-domain scenarios.

Stakeholder-centric explanations for black-box decisions: an XAI process model and its application to automotive goodwill assessments.

Machine learning has made tremendous progress in predictive performance in recent years. Despite these advances, employing machine learning models in high-stake domains remains challenging due to the opaqueness of many high-performance models. If their behavior cannot be analyzed, this likely decreases the trust in such models and hinders the acceptance of human decision-makers. Motivated by these challenges, we propose a process model for developing and evaluating explainable decision support systems that are tailored to the needs of different stakeholders. To demonstrate its usefulness, we apply the process model to a real-world application in an enterprise context. The goal is to increase the acceptance of an existing black-box model developed at a car manufacturer for supporting manual goodwill assessments. Following the proposed process, we conduct two quantitative surveys targeted at the application's stakeholders. Our study reveals that textual explanations based on local feature importance best fit the needs of the stakeholders in the considered use case. Specifically, our results show that all stakeholders, including business specialists, goodwill assessors, and technical IT experts, agree that such explanations significantly increase their trust in the decision support system. Furthermore, our technical evaluation confirms the faithfulness and stability of the selected explanation method. These practical findings demonstrate the potential of our process model to facilitate the successful deployment of machine learning models in enterprise settings. The results emphasize the importance of developing explanations that are tailored to the specific needs and expectations of diverse stakeholders.

On complexity of modeling the business process lifecycle

Business Process LifeCycle (BPLC) is a topic never disappearing from the discussions in the Business Process Management (BPM) research. Nevertheless, most graphical models of BPLC are simple and need extended textual explanations to clarify them. What is more, to the best of our knowledge, these models are separated from the target process that is being changed in the BPLC. The paper investigates a possibility to have a model where lifecycle activities are connected to the elements of the target process, and assess the practical usefulness of such a model. The research uses the Fractal Enterprise Modeling (FEM) technique for building the model. The paper presents a generic model of BPLC and its extension for one step, and discusses potential usefulness of the results achieved so far.

Творче конструювання особистості в автобіографічній повісті (на матеріалі творів «Хризантеми» Уляни Кравченко та «Десь на дні мого серця» Василя Лопати)

The article analyzes the tools of creative personality construction in two autobiographical novels, Chrysanthemums by by Ulyana Kravchenko and Somewhere at the Bottom of My Heart by Vasyl Lopata, and the thematic and narrative features of these works, as well as the importance of including documentary elements in the plot, are clarified. It is determined that the tools that helped to depict the formation of the author's creative personality in an autobiographical story were a distanced view of oneself, both in appearance and in the inner world, the use of humor, and emphasis on important stages. There is a difference in the narrative style – Ulyana Kravchenko focuses on the goal of developing intellectually since childhood, evaluates herself according to her readiness to serve the people, to form herself according to artistic laws, so the plot is often interrupted by lyrical inserts and reflections, although she simultaneously supplements it with historical and portrait essays. He compensates for his silence and loneliness in childhood by realizing it in music and his first attempts at writing, and explains the lack of words already when writing his memoirs by his own emotionality, creating an aura of mystery and understatement. V. Lopata introduces the main narrator into the story for a different purpose: to provide textual explanations, comment on the narrator-child's story, and express a short future continuation of the described events, the author much less often focuses on the process of recollection itself, having clearly defined for himself its purpose: to find out the origins of his own creative formation, which has already taken place (Ulyana Kravchenko writes about it as a process). The fragment of the diary, introduced into the textual space of the work, demonstrates the naivety and vulnerability of the protagonist in the conditions of orphanhood and war, and shows the boy's tendency to perceive even everyday events deeply in order to later convey them in his work. Through the adult narrator, the childhood traumas are closed and «liberation» from the described situations of childhood powerlessness is projected into the future.

Ordinal motifs in lattices

Lattices are a commonly used structure for the representation and analysis of relational and ontological knowledge. In particular, the analysis of these requires a decomposition of a large and high-dimensional lattice into a set of understandably large parts. With the present work we propose /ordinal motifs/ as analytical units of meaning. We study these ordinal substructures (or standard scales) through order-embeddings and (full) scale-measures of formal contexts from the field of formal concept analysis. We show that the underlying decision problems are NP-complete and provide results on how one can incrementally identify ordinal motifs to save computational effort. Accompanying our theoretical results, we demonstrate how ordinal motifs can be leveraged to achieve textual explanations based on principles from human computer interaction.

Explainable AI for Medical Data: Current Methods, Limitations, and Future Directions

With the power of parallel processing, large datasets,and fast computational resources, deep neural networks (DNNs) have outperformed highly trained and experienced human experts in medical applications. However, the large global community of healthcare professionals, many of whom routinely face potentially life-or-death outcomes with complex medicolegal consequences, have yet to embrace this powerful technology. The major problem is that most current AI solutions function as a metaphorical black-box positioned between input data and output decisions without a rigorous explanation for their internal processes. With the goal of enhancing trust and improving acceptance of AI-based technology in clinical medicine, there is a large and growing effort to address this challenge using eXplainable AI (XAI), a set of techniques, strategies, and algorithms with an explicit focus on explaining the “hows and whys” of DNNs. Here, we provide a comprehensive review of the state-of-the-art XAI techniques concerning healthcare applications and discuss current challenges and future directions. We emphasize the strengths and limitations of each category, including image, tabular, and textual explanations, and explore a range of evaluation metrics for assessing the effectiveness of XAI solutions. Finally, we highlight promising opportunities for XAI research to enhance the acceptance of DNNs by the healthcare community.

Design Heuristics Cards for circular economy to support generating ideas

The field of design for circular economy lacks comprehensive and systematic design aids to support users to generate ideas quickly and effectively. This study uses ReSOVLE as a framework, incorporates the lifecycle design approaches, and constructs the Circular Economy Design Heuristics Cards (CEDHC) by extracting strategies relevant to innovative product design, service design, and business model innovation into the form of design heuristics to support ideation, and incorporating TRIZ problem solving strategies to facilitate the iteration of original ideas. The CEDHC include design heuristics, textual explanations, cases and TRIZ 40 inventive principles. Two sessions were conducted to illustrate the efficacy of the CEDHC to support idea generation in the field of design for circular economy.

Design and development of a fuzzy explainable expert system for a diagnostic robot of COVID-19

<p><span lang="EN-US">Expert systems have been widely used in medicine to diagnose different diseases. However, these rule-based systems only explain why and how their outcomes are reached. The rules leading to those outcomes are also expressed in a machine language and confronted with the familiar problems of coverage and specificity. This fact prevents procuring expert systems with fully human-understandable explanations. Furthermore, early diagnosis involves a high degree of uncertainty and vagueness which constitutes another challenge to overcome in this study. This paper aims to design and develop a fuzzy explainable expert system for coronavirus disease-2019 (COVID-19) diagnosis that could be incorporated into medical robots. The proposed medical robotic application deduces the likelihood level of contracting COVID-19 from the entered symptoms, the personal information, and the patient's activities. The proposal integrates fuzzy logic to deal with uncertainty and vagueness in diagnosis. Besides, it adopts a hybrid explainable artificial intelligence (XAI) technique to provide different explanation forms. In particular, the textual explanations are generated as rules expressed in a natural language while avoiding coverage and specificity problems. Therefore, the proposal could help overwhelmed hospitals during the epidemic propagation and avoid contamination using a solution with a high level of explicability.</span></p>

Healthcare 3.0: How to Transform Machine Learning Prototypes into Functional Healthcare Applications for Diagnostic Assistance?

AbstractBackgroundAs the number of elderly people is rapidly increasing, we are faced with the challenge that diagnostic services are demanded more frequently. At the same time, the number of medical centers and available experts remains almost constant. Tools for diagnostic assistance are urgently needed to improve the effectiveness of healthcare. In four externally funded projects, we will investigate aspects and strategies of how machine learning prototype systems can be translated into functional healthcare applications.MethodIn the ongoing project “ExplAInation” funded by the German research foundation (DFG), we are developing an artificial neural network framework to generate visual and textual explanations to improve the comprehensibility and interpretability of deep learning models. This effort includes participatory research with clinical users. In the complementary project “TESIComp” funded by the Federal Ministry of Education and Research (BMBF), we will investigate ethical and social aspects of the emerging field of computational psychiatry. Patients, caregivers, and medical doctors will be interviewed, which will be qualitatively analyzed. Within the German Medical Informatics Initiative, we will contribute to the project “Open Medical Inference” (OMI), which will develop a network of distributed machine learning evaluation services. We also participate in the international “Clinical AI‐based Diagnostics” (CAIDX) consortium, which receives funding from the European Interreg Baltic Sea Region program.ResultWe developed a deep learning application for the detection of dementia atrophy patterns in brain MRI scans. Derived relevance maps highlight diagnostically important brain areas for further evaluation by the radiologists. Interviews with clinicians will provide information on expectations, key requirements and the functional utility of machine learning‐based assistance. Interviews with patients and caregivers will elucidate future changes and challenges in the doctor’s role and responsibilities. The OMI network will allow hospitals to use distributed machine learning evaluation services remotely, without the need of operating all the tools locally. In CAIDX, we will develop best‐practice guidelines for the overarching process of integrating machine learning prototypes and commercial tools into the hospital.ConclusionIn “Healthcare 3.0”, the digital transformation will change current diagnostic procedures and roles. Our activities focus on the involved stakeholders, regulatory aspects and implementation strategies to better steer this process.