Human Users Research Articles

Developing interpretable models and techniques for explainable AI in decision-making

The rapid proliferation of artificial intelligence (AI) technologies across various industries and decision-making processes has undeniably transformed the way of approaching complex problems and tasks. AI systems have proven their prowess in areas such as healthcare, finance, and autonomous systems, revolutionizing how decisions are made. Nevertheless, this proliferation of AI has raised critical concerns regarding the transparency, accountability, and fairness of these systems, as many of the state-of-the-art AI models often resemble complex black boxes. These intricate models, particularly deep learning neural networks, harbor non-linear relationships that are difficult for human users to decipher, thereby raising concerns about bias, fairness, and overall trustworthiness in AI-driven decisions. The urgency of this issue is underscored by the realization that AI should not merely be accurate; it should also be interpretable. Explainable AI (XAI) has emerged as a vital field of research, emphasizing the development of models and techniques that render AI systems comprehensible and transparent in their decision-making processes. This paper investigates into the relevance and significance of XAI across various domains, including healthcare, finance, and autonomous systems, where the ability to understand the rationale behind AI decisions is paramount. In healthcare, where AI assists in diagnosis and treatment, the interpretability of AI models is crucial for clinicians to make informed decisions. In finance, applications like credit scoring and investment analysis demand transparent AI to ensure fairness and accountability. In the realm of autonomous systems, transparency is indispensable to guarantee safety and compliance with regulations. Moreover, government agencies in areas such as law enforcement and social services require interpretable AI to maintain ethical standards and accountability. This paper also highlights the diverse array of research efforts in the XAI domain, spanning from model-specific interpretability methods to more general approaches aimed at unveiling complex AI models. Interpretable models like decision trees and rule-based systems have gained attention for their inherent transparency, while integrating explanation layers into deep neural networks strives to balance accuracy with interpretability. The study emphasizes the significance of this burgeoning field in bridging the gap between AI's advanced capabilities and human users' need for comprehensible AI systems. It seeks to contribute to this field by exploring the design, development, and practical applications of interpretable AI models and techniques, with the ultimate goal of enhancing the trust and understanding of AI-driven decisions.

Stream Flow Management and Institutions: Studying Water Quality and Aquatic Ecosystems in Dry Seasons

Stream flow management during the dry season involves balancing human water needs with the environmental requirements of the river. The allocation of stream water flow is determined by considering 10-year frequency low flow as the total available water amount, with instream flow adopted to consider the flow requirements of the river environment. However, conflicts arise over water rights between human users and the river environment during water scarcity. This study conducted an empirical-based analysis of minimum flow for river environments, focusing on water quality and aquatic ecosystems. Additionally, the perceptions of stakeholders of securing and using stream flow were investigated through a questionnaire. The outcomes of this study can contribute to enhancing technical methods for calculating instream flow and improving systems for managing river water during dry seasons.

인공지능 기반의 문학번역에 관한 저작권법 문제

Translation is a form of creativity, and AI creativity is making remarkable strides in many fields, including music, art, and literature, but especially in literature translation, where AI-based machine translation is becoming a reality. The recent happening about the winner of the Newcomer Award Competition of Literature Translation Institute of Korea, who was found to have utilized machine translation, has drawn attention to the question of how to set the boundaries between machine and human translation. I argue that AI-generated translation (AIGT), which is a purely AI translation without human assistance, is only reproduction under copyright law, while AI-assisted translation (AIAT), which is a human translation with the help of AI, can be a derivative work under copyright law as a translation. The distinguishing factor between the two is the control and subjectivity of human users of machine translation in the post-editing process, and the context surrounding the translation. This paper points out that excessive pre-editing by human users to facilitate machine translation may undermine the diversity of the original text. Once the legal status of machine translation output has been determined, I discuss the legal responsibility of human users to exercise due care at the stage of publication. The legal liability of failing to disclose that a work is the result of machine translation is also discussed, as well as plagiarism liability, which is not a legal liability.

UJI COBA RANCANGAN MEDIA INTERAKTIF TERHADAP PENINGKATAN PENGETAHUAN TENTANG KARIES PADA ANAK SEKOLAH DASAR

Dental caries are dental and oral diseases that often occur in communities. Knowledge deficit is an indicator of a high incidence of dental caries. Media is a tool for delivering messages from sender to recipient, which can be used in the process of providing information to increase knowledge. Interactive media is a form of media that contains interactivity or a mutually active relationship between humans (users) and products (media). This study aimed to determine the results of a feasibility test of interactive media design for caries and trials of interactive media design for caries on student knowledge. This was a quantitative study with a pre-experimental design using a one-group pretest-posttest design. Before the trials of interactive media design to students' knowledge, preceded by an assessment of the feasibility test of the interactive media design by media experts. Samples were taken by total sampling with a total sample of 54 fifth grade students at SDIT Daarul Jihad. The results showed that the feasibility test for interactive media design get a feasibility percentage of 93.8% with a very feasible category. The result also showed there was an increase in knowledge of dental caries with an average pre-test score of 66.48 to 90.37 in the post-test average score the average increasing score 26,39. The Wilcoxon test showed a significance level of 0.000 < 0.05. This showed that there was an influence from the results of the trials interactive media design for caries on the increasing knowledge of fifth grade elementary school students.

Postphenomenology and education: From cyborg students to immersive classrooms

This analysis uses a postphenomenological lens to provide insight into the shift occurring within society at large. It focuses on the educational domain, and arguing for a reevaluation of instructive approaches. Philosophical research into technology and education is seemingly lacking, and so this article seeks to fill the present gaps. This analysis initially delves into the postphenomenological frameworks of technological mediation, intentionality, and dimensions, to clearly differentiate the embodiment and cybernetic relationships as they are understood within various texts. Following this the epistemic and practical dimensions of these relations are explored to then be juxtaposed with the descriptive argument of the overall cultivating cybernetic relationship between human user and technological artefact in contemporary times, using the smartphone as the core example case study. Finally, a normative argument is made considering the previous cybernetic insight, in that if the classroom setting is to evolve and adapt, it too must embed technology within the classroom, lessons, and overall educational engagement. It represents an equalizing technological balance of cybernetic student and immersive classroom where the intentionality of user, technology, and classroom blend together to continue the cultivation of our blossoming relationship with technology.

On the interpretability of part-prototype based classifiers: a human centric analysis

Part-prototype networks have recently become methods of interest as an interpretable alternative to many of the current black-box image classifiers. However, the interpretability of these methods from the perspective of human users has not been sufficiently explored. In addition, previous works have had major issues with following proper experiment design and task representation that limit their reliability and validity. In this work, we have devised a framework for evaluating the interpretability of part-prototype-based models from a human perspective that solves these issues. The proposed framework consists of three actionable metrics and experiments. The results of these experiments will reveal important and reliable interpretability related properties of such models. To demonstrate the usefulness of our framework, we performed an extensive set of experiments using Amazon Mechanical Turk. They not only show the capability of our framework in assessing the interpretability of various part-prototype-based models, but they also are, to the best of our knowledge, the most comprehensive work on evaluating such methods in a unified framework.

Redefining User Expectations: The Impact of Adjustable Social Autonomy in Human–Robot Interaction

To promote the acceptance of robots in society, it is crucial to design systems exhibiting adaptive behavior. This is particularly needed in various social domains (e.g., cultural heritage, healthcare, education). Despite significant advancements in adaptability within Human-Robot Interaction and Social Robotics, research in these fields has overlooked the essential task of analyzing the robot’s cognitive processes and their implications for intelligent interaction (e.g., adaptive behavior, personalization). This study investigates human users’ satisfaction when interacting with a robot whose decision-making process is guided by a computational cognitive model integrating the principles of adjustable social autonomy. We designed a within-subjects experimental study in the domain of Cultural Heritage, where users (e.g., museum visitors) interacted with the humanoid robot Nao. The robot’s task was to provide the user with a museum exhibition to visit. The robot adopted the delegated task by exerting some degree of discretion, which required different levels of autonomy in the task adoption, relying on its capability to have a theory of mind. The results indicated that as the robot’s level of autonomy in task adoption increased, user satisfaction with the robot decreased, whereas their satisfaction with the tour itself improved. Results highlight the potential of adjustable social autonomy as a paradigm for developing autonomous adaptive social robots that can improve user experiences in multiple HRI real domains.

A Biobanking System for Diagnostic Images: Architecture Development, COVID-19-Related Use Cases, and Performance Evaluation.

Systems capable of automating and enhancing the management of research and clinical data represent a significant contribution of information and communication technologies to health care. A recent advancement is the development of imaging biobanks, which are now enabling the collection and storage of diagnostic images, clinical reports, and demographic data to allow researchers identify associations between lifestyle and genetic factors and imaging-derived phenotypes. The aim of this study was to design and evaluate the system performance of a network for an operating biobank of diagnostic images, the Bio Check Up Srl (BCU) Imaging Biobank, based on the Extensible Neuroimaging Archive Toolkit open-source platform. Three usage cases were designed focusing on evaluation of the memory and computing consumption during imaging collections upload and during interactions between two kinds of users (researchers and radiologists) who inspect chest computed tomography scans of a COVID-19 cohort. The experiments considered three network setups: (1) a local area network, (2) virtual private network, and (3) wide area network. The experimental setup recorded the activity of a human user interacting with the biobank system, which was continuously replayed multiple times. Several metrics were extracted from network traffic traces and server logs captured during the activity replay. Regarding the diagnostic data transfer, two types of containers were considered: the Web and the Database containers. The Web appeared to be the more memory-hungry container with a higher computational load (average 2.7 GB of RAM) compared to that of the database. With respect to user access, both users demonstrated the same network performance level, although higher resource consumption was registered for two different actions: DOWNLOAD & LOGOUT (100%) for the researcher and OPEN VIEWER (20%-50%) for the radiologist. This analysis shows that the current setup of BCU Imaging Biobank is well provisioned for satisfying the planned number of concurrent users. More importantly, this study further highlights and quantifies the resource demands of specific user actions, providing a guideline for planning, setting up, and using an image biobanking system.

Conversational Breakdown Detector for a Motivational Interviewing Conversational Agent

A conversational breakdown in human-chatbot interaction refers to a disruption or failure in the communicative flow between the human user and the chatbot. To recover a disrupted conversation, the first step is to detect the breakdown. Researchers have proposed methods using supervised learning and semi-supervised learning in dialogue systems to achieve the goal of detecting conversational breakdown. However, few studies have focused on detecting breakdowns in automated therapeutic conversations, especially conversations led by motivational interviewing chatbots. The presence of conversational breakdowns has negative impacts on the human-chatbot interaction, such as frustration, dissatisfaction, or loss of trust. This gap suggests a need to build a robust and efficient conversational breakdown detector that recognizes interruptions during the conversation. Conversational breakdown detection paves the way for further action to recover conversations. In this paper, I develop a novel, unifying framework called “CIMIC” for characterizing the conversational breakdowns of “MIBot,” a motivational interviewing conversational agent for smoking cessation. I collect 200 pieces of conversational data through Prolific and annotate them using the CIMIC framework with a group of four trained annotators. The annotated dataset is then applied as the training set to fine-tune GPT-3 models to build a conversational breakdown detector for the MIBot.

Is Someone There or Is That the TV? Detecting Social Presence Using Sound

Social robots in the home will need to solve audio identification problems to better interact with their users. This article focuses on the classification between (a) natural conversation that includes at least one co-located user and (b) media that is playing from electronic sources and does not require a social response, such as television shows. This classification can help social robots detect a user’s social presence using sound. Social robots that are able to solve this problem can apply this information to assist them in making decisions, such as determining when and how to appropriately engage human users. We compiled a dataset from a variety of acoustic environments that contained either natural or media audio, including audio that we recorded in our own homes. Using this dataset, we performed an experimental evaluation on a range of traditional machine learning classifiers and assessed the classifiers’ abilities to generalize to new recordings, acoustic conditions, and environments. We conclude that a C-Support Vector Classification (SVC) algorithm outperformed other classifiers. Finally, we present a classification pipeline that in-home robots can utilize, and we discuss the timing and size of the trained classifiers as well as privacy and ethics considerations.

A Neural Network-Based Random Access Protocol for Crowded Massive MIMO Systems.

Fifth-generation (5G) and beyond networks are expected to serve large numbers of user equipments (UEs). Grant-based random access (RA) protocols are efficient when serving human users, typically with large data volumes to transmit. The strongest user collision resolution (SUCRe) is the first protocol that effectively uses the many antennas at the 5G base station (BS) to improve connectivity performance. In this paper, our proposal involves substituting the retransmission rule of the SUCRe protocol with a neural network (NN) to enhance the identification of the strongest user and resolve collisions in a decentralized manner on the UEs' side. The proposed NN-based procedure is trained offline, admitting different congestion levels of the system, aiming to obtain a single setup able to operate with different numbers of UEs. The numerical results indicate that our method attains substantial connectivity performance improvements compared to other protocols without requiring additional complexity or overhead. In addition, the proposed approach is robust regarding variations in the number of BS antennas and transmission power while improving energy efficiency by requiring fewer attempts on the RA stage.

AudioMNIST: Exploring Explainable Artificial Intelligence for audio analysis on a simple benchmark

Explainable Artificial Intelligence (XAI) is targeted at understanding how models perform feature selection and derive their classification decisions. This paper explores post-hoc explanations for deep neural networks in the audio domain. Notably, we present a novel Open Source audio dataset consisting of 30,000 audio samples of English spoken digits which we use for classification tasks on spoken digits and speakers’ biological sex. We use the popular XAI technique Layer-wise Relevance Propagation LRP to identify relevant features for two neural network architectures that process either waveform or spectrogram representations of the data. Based on the relevance scores obtained from LRP, hypotheses about the neural networks’ feature selection are derived and subsequently tested through systematic manipulations of the input data. Further, we take a step beyond visual explanations and introduce audible heatmaps. We demonstrate the superior interpretability of audible explanations over visual ones in a human user study.

Design guidelines for sustainable utilization of agricultural appropriate technology: Enhancing human factors and user experience

Abstract The utilization of appropriate technology (AT) has become the Indonesian government strategy to increase productivity of agricultural commodities due to its simplicity and cost-effectiveness. However, the current utilization of AT remains suboptimal mostly due to design deficiencies that insufficiently account for human factors and user experience. In response, the aim of this study is to establish comprehensive AT design guidelines for sustainable AT utilization, focused on agricultural processing machines. An intensive observation was initially conducted in a rural community in Indonesia, to summarize difficulties faced by AT users. Supported by an exhaustive review of literature, a total of 44 human factors related design criteria were defined. Subsequently, these criteria underwent rigorous validation through a questionnaire administered to 197 respondents, consisting of AT designers, experts, and users. Employing the framework of principal component analysis (PCA), novel dimensions of AT design criteria were suggested, encompassing safety and error prevention, functionality and economics, user-friendly, low physical effort, physical workspace compatibility, and perceptible information. To augment the insights gleaned from the PCA, a matrix of importance-performance analysis was created, affording a map of the relative significance and concurrent performance of the defined criteria. The implications of this study are further discussed.

Explainable empirical risk minimization

AbstractThe successful application of machine learning (ML) methods increasingly depends on their interpretability or explainability. Designing explainable ML (XML) systems is instrumental for ensuring transparency of automated decision-making that targets humans. The explainability of ML methods is also an essential ingredient for trustworthy artificial intelligence. A key challenge in ensuring explainability is its dependence on the specific human end user of an ML system. The users of ML methods might have vastly different background knowledge about ML principles, with some having formal training in the specific field and others having none. We use information-theoretic concepts to develop a novel measure for the subjective explainability of predictions delivered by a ML method. We construct this measure via the conditional entropy of predictions, given the user signal. Our approach allows for a wide range of user signals, ranging from responses to surveys to biophysical measurements. We use this measure of subjective explainability as a regularizer for model training. The resulting explainable empirical risk minimization (EERM) principle strives to balance subjective explainability and risk. The EERM principle is flexible and can be combined with arbitrary ML models. We present several practical implementations of EERM for linear models and decision trees. Numerical experiments demonstrate the application of EERM to weather prediction and detecting inappropriate language in social media.

Performance of Large Language Models on a Neurology Board–Style Examination

Recent advancements in large language models (LLMs) have shown potential in a wide array of applications, including health care. While LLMs showed heterogeneous results across specialized medical board examinations, the performance of these models in neurology board examinations remains unexplored. To assess the performance of LLMs on neurology board-style examinations. This cross-sectional study was conducted between May 17 and May 31, 2023. The evaluation utilized a question bank approved by the American Board of Psychiatry and Neurology and was validated with a small question cohort by the European Board for Neurology. All questions were categorized into lower-order (recall, understanding) and higher-order (apply, analyze, synthesize) questions based on the Bloom taxonomy for learning and assessment. Performance by LLM ChatGPT versions 3.5 (LLM 1) and 4 (LLM 2) was assessed in relation to overall scores, question type, and topics, along with the confidence level and reproducibility of answers. Overall percentage scores of 2 LLMs. LLM 2 significantly outperformed LLM 1 by correctly answering 1662 of 1956 questions (85.0%) vs 1306 questions (66.8%) for LLM 1. Notably, LLM 2's performance was greater than the mean human score of 73.8%, effectively achieving near-passing and passing grades in the neurology board examination. LLM 2 outperformed human users in behavioral, cognitive, and psychological-related questions and demonstrated superior performance to LLM 1 in 6 categories. Both LLMs performed better on lower-order than higher-order questions, with LLM 2 excelling in both lower-order and higher-order questions. Both models consistently used confident language, even when providing incorrect answers. Reproducible answers of both LLMs were associated with a higher percentage of correct answers than inconsistent answers. Despite the absence of neurology-specific training, LLM 2 demonstrated commendable performance, whereas LLM 1 performed slightly below the human average. While higher-order cognitive tasks were more challenging for both models, LLM 2's results were equivalent to passing grades in specialized neurology examinations. These findings suggest that LLMs could have significant applications in clinical neurology and health care with further refinements.

A mouse model of oral contraceptive exposure: Depression, motivation, and the stress response

Hormonal contraceptives, including oral contraceptives (OCs), regulate hormonal cycles and broadly affect physiological processes, including stress responsivity. Whereas many users describe overall improved mood, up to 10 % of OC users experience adverse effects, including depression and anxiety. Given the link between regulation of hypothalamic-pituitary-adrenal (HPA) axis, stress exposure, and risk for depression, it is likely that OC-effects on stress mediate increased risk or increased resilience to these disorders. In this study, we developed and characterized a tractable mouse model of OC exposure with which to identify the mechanisms underlying OC modulation of brain, behavior, and mood. Specifically, we aimed to determine whether translationally relevant doses of OC-hormones in mice mimic changes in stress responsivity observed in humans taking OCs and describe behavioral changes during OC exposure. Young adult female C57Bl/6 N mice received daily ethinyl estradiol (EE) and levonorgestrel (LVNG) in 10 % sucrose, EE and drospirenone (DRSP) in 10 % sucrose, or 10 % sucrose alone. Translationally relevant doses of EE + LVNG-exposure, but not EE + DRSP, suppressed the acute stress response, consistent with effects observed in human OC users. EE + LVNG caused a specific anhedonia-like effect, without broad changes in stress-coping behavior, other depression-like behaviors, or anxiety-like behaviors. The suppression of regular estrous cycling, together with the blunting of the corticosterone response to acute stress, demonstrate the utility of this model for future studies to identify the mechanisms underlying OC interactions with stress, motivation, and risk for depression.

Tracking and Detecting Depression Level using Facial Recognition and PEN&IQ Test

These Human facial expressions convey a lot of information visually rather than articulately. Facial expression recognition plays a crucial role in the area of human-machine interaction. Automatic facial expression recognition system has many applications including, but not limited to, human behaviour understanding, detection of mental disorders, and synthetic human expressions. Recognition of facial expression by computer with high recognition rate is still a challenging task. Two popular methods utilized mostly in the literature for the automatic FER systems are based on geometry and appearance. Facial Expression Recognition usually performed in four-stages consisting of pre-processing, face detection, feature extraction, and expression classification. The human face is an important part of an individual’s body and plays an important role in knowing the individual’s mood. The face is where a human expresses all his basic emotions. In the existing system, they examine the mental state manually by assessing them but have many disadvantages like we cannot predict any accurate solutions based on the assessment score because we might be not sure what kind of emotions the human user would be all time. To overcome this problem, a novel system is proposed to suggest an effective solution for predicting the mental state dynamically, we propose a hybrid architecture invoking facial based emotion sequence, PEN test and IQ test. By consistent monitoring of a human’s emotion and subjecting to PEN and IQ tests, the human’s mental state is routed. Combination of the above three techniques provides promising results for mental state and self-control

Characteristics of Rear-End Collisions: A Comparison between Automated Driving System-Involved Crashes and Advanced Driving Assistance System-Involved Crashes

With the increasing number of vehicles equipped with automated driving systems (ADS) being tested on public roads and the expanding market share of vehicles equipped with advanced driving assistance systems (ADAS), the number of ADS- or ADAS-involved crashes increases. Thus, it is necessary to investigate the distribution of ADS- and ADAS-involved crashes and the factors leading to them. The rear-end collision has been found to dominate ADS-involved crashes. However, no research has explored the conditions when ADS-involved rear-end collisions are more likely to happen and no research has investigated ADAS-involved rear-end crashes. Based on 130 ADS-involved crashes and 84 ADAS-involved crashes extracted from a dataset collected by the National Highway Traffic Safety Administration (NHTSA) between July 2021 and May 2022, this study explored the crash patterns, especially rear-end crashes, of ADS- and ADAS-controlled vehicles. Results show that rear-end collisions dominate both ADS- and ADAS-involved crashes, especially ADAS-involved crashes. The type of both ADS-involved and ADAS-involved crashes was affected by the speed of the ego-vehicle relative to the posted speed limit. Further, the type of ADS-involved crash was affected by the pre-crash movement of the crash partner, while the type of ADAS-involved crash was further associated with the road type. The findings can provide insights into the design of ADAS and ADS control algorithms, the external human-machine interface design of the vehicles with ADS or ADAS, and the training program of human road users to improve traffic safety in mixed traffic.

Social Interaction‐Aware Dynamical Models and Decision‐Making for Autonomous Vehicles

Interaction‐aware autonomous driving (IAAD) is a rapidly growing field of research that focuses on the development of autonomous vehicles (AVs) that are capable of interacting safely and efficiently with human road users. This is a challenging task, as it requires the AV to be able to understand and predict the behaviour of human road users. In this literature review, the current state of IAAD research is surveyed. Commencing with an examination of terminology, attention is drawn to challenges and existing models employed for modeling the behaviour of drivers and pedestrians. Next, a comprehensive review is conducted on various techniques proposed for interaction modeling, encompassing cognitive methods, machine‐learning approaches, and game‐theoretic methods. The conclusion is reached through a discussion of potential advantages and risks associated with IAAD, along with the illumination of pivotal research inquiries necessitating future exploration.

Language-Model-Based Paired Variational Autoencoders for Robotic Language Learning

Human infants learn language while interacting with their environment in which their caregivers may describe the objects and actions they perform. Similar to human infants, artificial agents can learn language while interacting with their environment. In this work, first, we present a neural model that bidirectionally binds robot actions and their language descriptions in a simple object manipulation scenario. Building on our previous Paired Variational Autoencoders (PVAE) model, we demonstrate the superiority of the variational autoencoder over standard autoencoders by experimenting with cubes of different colours, and by enabling the production of alternative vocabularies. Additional experiments show that the model's channel-separated visual feature extraction module can cope with objects of different shapes. Next, we introduce PVAE-BERT, which equips the model with a pretrained large-scale language model, i.e., Bidirectional Encoder Representations from Transformers (BERT), enabling the model to go beyond comprehending only the predefined descriptions that the network has been trained on; the recognition of action descriptions generalises to unconstrained natural language as the model becomes capable of understanding unlimited variations of the same descriptions. Our experiments suggest that using a pretrained language model as the language encoder allows our approach to scale up for real-world scenarios with instructions from human users.