Research in human-centered explainable AI has increasingly emphasized personalization and adaptation, highlighting the need to tailor explanations to individual users and their specific contexts. Prior work suggests that explanation demand (that is, whether users want explanations) varies across tasks and users, motivating the development of adaptive explanation systems that deliver explanations at appropriate moments. However, it remains unclear how explanation demand should be modeled in such systems.

The emergence of large language models (LLMs) has made it easier than ever to create chatbots capable of generating human-like responses to user inputs. Moreover, improvements in text-to-speech and speech-to-text make it possible to converse with these systems, not just in text but also through speech. These improvements have led to an increase in chatbot applications across various contexts, such as customer service and healthcare. This study examined the tendency to anthropomorphize chatbots in a mental health assessment context.

Socially assistive humanoid robots are becoming increasingly integrated into home environments, where they are expected to interact naturally and transparently with users. In this context, the aim of the study presented in this paper is to understand how they should deliver explanations, especially when the robots cannot fulfill a user’s request. We explore different explanation modalities by combining spoken explanations with an additional element (speech alone, speech plus lights, speech plus sounds, and speech plus gestures) and examining user preferences among them. We designed a video-based between-subjects user study featuring the Nao robot across three everyday scenarios where the robot fails to perform a task due to overheated motors. Participants evaluated four explanation modalities and provided feedback on their preferences. Our findings show that multimodal explanations are generally favored over speech alone, with speech and lights being the most preferred. Although scenario context did not have a statistically significant association with modality preference, qualitative feedback highlights the importance of context-aware and adaptive communication strategies. These insights provide practical guidance for designing more natural and effective human-robot interactions.

Automatic checkout systems are designed to predict a complete shopping receipt using an image from the checkout area. These systems require high classification accuracy across numerous classes and must operate in real-time, despite domain differences between training data and real-world conditions. Building on recent advancements, we propose a method that outperforms current solutions and can be applied in real-time in automatic checkout systems. Our method leverages the Segment Anything Model to extract high-quality masks from lab product images, which are then transformed into synthetic checkout images and adapted to the real domain using contrastive unpaired translation. We train a product recognition model with data augmentation, named SCA+Y8, and further improve it through fine-tuning with pseudo-labels from unlabeled checkout images, resulting in an improved model called SCAFT+Y8. SCAFT+Y8 achieves a great increase in state-of-the-art performance, with an average receipt classification accuracy of 97.58%, and shows strong performance in smaller models, indicating the potential for deployment on low-cost edge devices.

Abstract Automatic recognition of grocery products can be used to improve customer flow at checkouts and reduce labor costs and store losses. Product recognition is, however, a challenging task for machine learning-based solutions due to the large number of products and their variations in appearance. In this work, we tackle the challenge of fine-grained product recognition by first extracting a large dataset from a grocery store containing products that are only differentiable by subtle details. Then, we propose a multimodal product recognition approach that uses product images with extracted OCR text from packages to improve fine-grained recognition of grocery products. We evaluate several image and text models separately and then combine them using different multimodal models of varying complexities. The results show that image and textual information complement each other in multimodal models and enable a classifier with greater recognition performance than unimodal models, especially when the number of training samples is limited. Therefore, this approach is suitable for many different scenarios in which product recognition is used to further improve recognition performance. The dataset can be found at https://github.com/Tubbias/finegrainocr .

Explanations in artificial intelligence (AI) ensure that users of complex AI systems understand why the system behaves as it does. Expectations that users may have about the system behaviour play a role since they co-determine appropriate content of the explanations. In this paper, we investigate user-desired content of explanations when the system behaves in unexpected ways. Specifically, we presented participants with various scenarios involving an automated text classifier and then asked them to indicate their preferred explanation in each scenario. One group of participants chose the type of explanation from a multiple-choice questionnaire, the other had to answer using free text.

We are largely used to hearing explanations. For example, if someone thinks you are sad today, they might reply to your “why?” with “because you were so Hmmmmm-mmm-mmm”. Today’s Artificial Intelligence (AI), however, is – if at all – largely providing explanations of decisions in a visual or textual manner. While such approaches are good for communication via visual media such as in research papers or screens of intelligent devices, they may not always be the best way to explain; especially when the end user is not an expert. In particular, when the AI’s task is about Audio Intelligence, visual explanations appear less intuitive than audible, sonified ones. Sonification has also great potential for explainable AI (XAI) in systems that deal with non-audio data – for example, because it does not require visual contact or active attention of a user. Hence, sonified explanations of AI decisions face a challenging, yet highly promising and pioneering task. That involves incorporating innovative XAI algorithms to allow pointing back at the learning data responsible for decisions made by an AI, and to include decomposition of the data to identify salient aspects. It further aims to identify the components of the preprocessing, feature representation, and learnt attention patterns that are responsible for the decisions. Finally, it targets decision-making at the model-level, to provide a holistic explanation of the chain of processing in typical pattern recognition problems from end-to-end. Sonified AI explanations will need to unite methods for sonification of the identified aspects that benefit decisions, decomposition and recomposition of audio to sonify which parts in the audio were responsible for the decision, and rendering attention patterns and salient feature representations audible. Benchmarking sonified XAI is challenging, as it will require a comparison against a backdrop of existing, state-of-the-art visual and textual alternatives, as well as synergistic complementation of all modalities in user evaluations. Sonified AI explanations will need to target different user groups to allow personalisation of the sonification experience for different user needs, to lead to a major breakthrough in comprehensibility of AI via hearing how decisions are made, hence supporting tomorrow’s humane AI’s trustability. Here, we introduce and motivate the general idea, and provide accompanying considerations including milestones of realisation of sonifed XAI and foreseeable risks.

Research in the social sciences has shown that expectations are an important factor in explanations as used between humans: rather than explaining the cause of an event per se, the explainer will often address another event that did not occur but that the explainee might have expected. For AI-powered systems, this finding suggests that explanation-generating systems may need to identify such end user expectations. In general, this is a challenging task, not the least because users often keep them implicit; there is thus a need to investigate the importance of such an ability. In this paper, we report an empirical study with 181 participants who were shown outputs from a text classifier system along with an explanation of why the system chose a particular class for each text. Explanations were both factual, explaining why the system produced a certain output or counterfactual, explaining why the system produced one output instead of another. Our main hypothesis was explanations should align with end user expectations; that is, a factual explanation should be given when the system's output is in line with end user expectations, and a counterfactual explanation when it is not. We find that factual explanations are indeed appropriate when expectations and output match. When they do not, neither factual nor counterfactual explanations appear appropriate, although we do find indications that our counterfactual explanations contained at least some necessary elements. Overall, this suggests that it is important for systems that create explanations of AI systems to infer what outputs the end user expected so that factual explanations can be generated at the appropriate moments. At the same time, this information is, by itself, not sufficient to also create appropriate explanations when the output and user expectations do not match. This is somewhat surprising given investigations of explanations in the social sciences, and will need more scrutiny in future studies.

Rapid technical advancements have led to dramatically improved abilities for artificial agents, and thus opened up for new ways of cooperation between humans and them, from disembodied agents such as Siris to virtual avatars, robot companions, and autonomous vehicles. It is therefore relevant to study not only how to maintain appropriate cooperation, but also where the responsibility for this resides and/or may be affected. While there are previous organisations and categorisations of agents and HAI research into taxonomies, situations with highly responsible artificial agents are rarely covered. Here, we propose a way to categorise agents in terms of such responsibility and agent autonomy, which covers the range of cooperation from humans getting help from agents to humans providing help for the agents. In the resulting diagram presented in this paper, it is possible to relate different kinds of agents with other taxonomies and typical properties. A particular advantage of this taxonomy is that it highlights under what conditions certain effects known to modulate the relationship between agents (such as the protégé effect or the "we"-feeling) arise.

Visual Analytics has brought forward many solutions to different tasks such as exploring topics, understanding user and customer behavior, comparing genomes, or detecting anomalies. Many of these solutions, if not most, are standalone applications with technological contributions which cannot be easily taken for: reuse in other domains, further improvement, benchmarking, or integration and deployment alongside other solutions. The latter can prove specially helpful for exploratory data analysis. This often leads researchers to re-implement solutions and thus to a suboptimal use of skills and resources. This paper discusses further the lack of off-the-shelf libraries for Visual Analytics, and proposes the creation of pluggable libraries on top of existing technologies such as Spark and Zeppelin. We provide an illustrative example of a pluggable, Visual Analytics library using these technologies.

Abstract To find correlations and cause and effect relationships in multivariate data sets is central in many data analysis problems. A common way of representing causal relations among variables is to use node‐link diagrams, where nodes depict variables and edges show relationships between them. When performing a causal analysis, analysts may be biased by the position of collected evidences, especially when they are at the top of a list. This is of crucial importance since finding a root cause or a derived effect, and searching for causal chains of inferences are essential analytic tasks when investigating causal relationships. In this paper, we examine whether sequential ordering influences understanding of indirect causal relationships and whether it improves readability of multi‐attribute causal diagrams. Moreover, we see how people reason to identify a root cause or a derived effect. The results of our design study show that sequential ordering does not play a crucial role when analyzing causal relationships, but many connections from/to a variable and higher strength/certainty values may influence the process of finding a root cause and a derived effect.

The analysis of large amounts of multidimensional road traffic data for anomaly detection is a complex task. Visual analytics can bridge the gap between computational and human approaches to detecting anomalous behavior in road traffic, making the data analysis process more transparent. In this paper, we present a visual analytics framework that provides support for: 1) the exploration of multidimensional road traffic data; 2) the analysis of normal behavioral models built from data; 3) the detection of anomalous events; and 4) the explanation of anomalous events. We illustrate the use of this framework with examples from a large database of real road traffic data collected from several areas in Europe. Finally, we report on feedback provided by expert analysts from Volvo Group Trucks Technology, regarding its design and usability.

In this article, we present an interactive prototype that aids the interpretation of large-scale gene expression data, showing how visualization techniques can be applied to support knowledge extraction from large datasets. The developed prototype was evaluated on a dataset of human embryonic stem cell-derived cardiomyocytes. The visualization approach presented here supports the analyst in finding genes with high similarity or dissimilarity across different experimental groups. By using an external overview in combination with filter windows, and various color scales for showing the degree of similarity, our interactive visual prototype is able to intuitively guide the exploration processes over the large amount of gene expression data.

This paper presents an empirical study that evaluates the effects of visualizing missing data on decision-making tasks. A comparison between three visualization techniques: (1) emptiness, (2) fuzziness, and (3) emptiness plus explanation, revealed that the latter technique induced significantly higher degree of decision-confidence than the visualization technique fuzziness. Moreover, emptiness plus explanation yield the highest number of risky choices of the three. This result suggests that uncertainty visualization techniques affect the decision-maker and the decisionconfidence. Additionally, the results indicate a possible relation between the degree of decision-confidence and the decision-maker's displayed risk behavior.

Monitoring dynamic objects in surveillance applications is normally a demanding activity for operators, not only because of the complexity and high dimensionality of the data but also because of other factors like time constraints and uncertainty. Timely detection of anomalous objects or situations that need further investigation may reduce operators’ cognitive load. Surveillance applications may include anomaly detection capabilities, but their use is not widespread, as they usually generate a high number of false alarms, they do not provide appropriate cognitive support for operators, and their outcomes can be difficult to comprehend and trust. Visual analytics can bridge the gap between computational and human approaches to detecting anomalous behavior in traffic data, making this process more transparent. As a step toward this goal of transparency, this article presents an evaluation that assesses whether visualizations of normal behavioral models of vessel traffic support two of the main analytical tasks specified during our field work in maritime control centers. The evaluation combines quantitative and qualitative usability assessments. The quantitative evaluation, which was carried out with a proof-of-concept prototype, reveals that participants who used the visualization of normal behavioral models outperformed the group that did not do so. The qualitative assessment shows that domain experts have a positive attitude toward the provision of automatic support and the visualization of normal behavioral models, as these aids may reduce reaction time and increase trust in and comprehensibility of the system.

To investigate the impact of visualizing car uncertainty on drivers' trust during an automated driving scenario, a simulator study was conducted. A between-group design experiment with 59 Swedish drivers was carried out where a continuous representation of the uncertainty of the car's ability to autonomously drive during snow conditions was displayed to one of the groups, whereas omitted for the control group. The results show that, on average, the group of drivers who were provided with the uncertainty representation took control of the car faster when needed, while they were, at the same time, the ones who spent more time looking at other things than on the road ahead. Thus, drivers provided with the uncertainty information could, to a higher degree, perform tasks other than driving without compromising with driving safety. The analysis of trust shows that the participants who were provided with the uncertainty information trusted the automated system less than those who did not receive such information, which indicates a more proper trust calibration than in the control group.

Despite the growing number of systems providing visual analytic support for investigative analysis, few empirical studies include investigations on the analytical reasoning process that needs to be supported. In this paper, we present an approach to evaluate the ability of certain visual representations from an integrated visual-computational environment to support the completion of representative tasks. The problem area studied is the detection and identification of anomalous vessels and situations while monitoring maritime traffic data. This paper presents: (1) a brief review of current evaluation methodologies within information visualization and visual analytics, (2) an analysis of operator's analytical reasoning process (derived from field work in maritime control centers and a literature review on analytical reasoning theories), (3) a list of representative tasks for usability evaluation and (4) an approach to evaluate the use of normal behavioral models representations during the detection process.

The surveillance of large sea areas often generates huge amounts of multidimensional data. Exploring, analyzing and finding anomalous behavior within this data is a complex task. Confident decisions upon the abnormality of a particular vessel behavior require a certain level of situation awareness that may be difficult to achieve when the operator is overloaded by the available information. Based on a visual analytics process model, we present a novel system that supports the acquisition of situation awareness and the involvement of the user in the anomaly detection process using two layers of interactive visualizations. The system uses an interactive data mining module that supports the insertion of the user's knowledge and experience in the creation, validation and continuous update of the normal model of the environment.

This paper highlights the importance of uncertainty visualization in information fusion, reviews general methods of representing uncertainty and presents perceptual and cognitive principles from Tufte, Chambers and Bertin as well as users experiments documented in the literature. Examples of uncertainty representations in information fusion are analyzed using these general theories. These principles can be used in future theoretical evaluations of existing or newly developed uncertainty visualization techniques before usability testing with actual users.

This paper focuses on the capability of artificial evolution to produce tool use behaviors of different complexity in simulated robotic agents and in the absence of learning or other lifetime methods. The results show by example that tool use behaviors of different complexity can evolve and do not necessarily rely on reasoning abilities