📚 Research Digest

Finding: This paper presents a comprehensive methodological framework (HCAI-MF) for implementing human-centered AI, featuring five key components including requirement hierarchies, method taxonomies, processes, interdisciplinary collaboration approaches, and multilevel design paradigms to guide practitioners in developing AI systems that prioritize human needs.

Why it matters: This directly addresses how to systematically apply human-centered design principles to AI development, providing concrete methodological guidance for HCI practitioners working on AI systems and human-AI collaboration.

Method: Uses a case study to demonstrate the framework's practical application and provides a "three-layer" implementation strategy with actionable recommendations.

Jargon: HCAI (Human-Centered Artificial Intelligence) - design philosophy prioritizing humans throughout AI system lifecycle from design to deployment and use.

Finding: Social fitness platforms like Strava boost motivation and community but create psychological risks, particularly for users prone to negative social comparisons and low self-compassion, with health motives and social recognition amplifying negative effects more than age factors.

Why it matters: This demonstrates how digital platform design choices around social comparison and gamification can have complex psychological effects that vary by user characteristics and motivations, directly informing behavioral science understanding of technology's impact on human behavior.

Method: Mixed-methods study combining quantitative survey of 114 Strava users with qualitative interviews of 4 users, examining social comparison tendencies and psychological outcomes.

Jargon: Social comparison orientation - tendency to evaluate oneself relative to others; self-compassion - being kind to oneself during difficulties or failures.

Finding: This empirical study compares usability and accessibility of scientific journal websites across three conditions: existing sites, human-designed prototypes following Universal Design principles, and AI-generated prototypes, using eye-tracking, surveys, and automated accessibility assessments.

Why it matters: Directly addresses how AI tools are changing interface design processes and provides concrete usability evaluation methods relevant to HCI education and understanding AI's impact on design work.

Method: Uses triangulated evaluation approach combining eye-tracking data, user surveys, and automated WCAG 2.1 AA compliance testing across multiple prototype conditions.

Jargon: Universal Design (UD) = design approach making products usable by all people; WCAG 2.1 AA = Web Content Accessibility Guidelines compliance level.

Finding: This study developed and validated design guidelines for tangible user interfaces (TUIs) specifically for children with dyslexia, showing significant improvements in letter recognition and spelling through the Tangibell prototype tested with 16 participants aged 7-12.

Why it matters: This provides concrete, evidence-based design principles for creating accessible learning technologies, directly informing HCI design practices for inclusive educational tools and demonstrating how physical interaction can enhance learning for specific populations.

Method: Mixed-methods approach combining expert validation from 7 HCI and design experts with experimental testing on children with dyslexia, measuring learning outcomes and gathering user preferences through questionnaires.

Jargon: TUIs (Tangible User Interfaces) - physical objects that users can manipulate to interact with digital systems; Universal Design - design approach making products accessible to people with widest range of abilities.

Finding: The system combines eye blink detection and hand gesture recognition using MediaPipe to create a touch-free HCI interface, achieving 94-96% accuracy for applications in assistive technology and sterile environments.

Why it matters: This demonstrates practical multimodal interface design that addresses accessibility and inclusive design principles, key considerations in modern HCI and UX design.

Method: Custom algorithm differentiates intentional slow blinks from natural ones, integrating facial and hand landmark tracking through Google's MediaPipe framework.

Jargon: MediaPipe - Google's framework for building multimodal applied ML pipelines; landmark tracking - identifying key points on faces/hands for gesture recognition.

Finding: Analysis of 5,000 LinkedIn job postings shows AI skills are now required in 27.8% of knowledge worker positions (376% growth since ChatGPT's release), with AI-trained workers earning 17.7% wage premiums, leading to a proposed three-dimensional performance evaluation model: AI Tool Mastery, Collaborative Work Quality, and Human-AI Synergy.

Why it matters: Directly addresses how AI tools like ChatGPT are reshaping work evaluation systems and human-AI collaboration, which is crucial for understanding the future of work and designing effective performance measurement systems.

Method: Systematic analysis of job advertisements and salary data from major employment platforms over a 2-year period spanning ChatGPT's release.

Jargon: Human-AI Synergy refers to the hybrid skills developed through effective collaboration between humans and AI systems.

Finding: This study reveals that student motivation in examination-oriented systems emerges from a negotiated balance between intrinsic needs (autonomy, competence, curiosity) and external pressures, with timely effort-focused feedback being crucial for sustaining engagement and self-regulation.

Why it matters: The findings provide actionable insights for designing learning environments that support intrinsic motivation while managing external pressures, directly informing goal-setting practices, feedback systems, and motivation-based interventions in educational contexts.

Method: Semi-structured interviews with 25 undergraduates analyzed through thematic analysis, integrating Self-Determination Theory with expectancy-value and attribution theories.

Jargon: Self-Determination Theory (SDT) - framework focusing on three basic psychological needs: autonomy (feeling volitional), competence (feeling effective), and relatedness (feeling connected); Expectancy-value theory - motivation depends on expectation of success and perceived value of the task.

Finding: The PXF framework generates exam questions using AI while incorporating pedagogical alignment (Bloom's Taxonomy), explainability for human understanding, and fairness through bias detection, achieving 91% accuracy and reducing question creation time by 84%.

Why it matters: This demonstrates how AI can augment educational practices while maintaining pedagogical rigor and addressing ethical concerns about bias, directly relevant to AI's impact on learning sciences and human-AI collaboration in education.

Method: Experimental validation using educational datasets with a modular AI architecture including pedagogy alignment, explainability engine, fairness module, and human-in-the-loop review interface.

Jargon: Bloom's Taxonomy - educational framework categorizing learning objectives from basic recall to complex evaluation and creation.

Finding: This narrative literature review identifies four evidence-based design dimensions for effective digital books in e-learning: instructional design foundations, purposeful multimedia integration based on cognitive principles, self-regulated learning support, and technological usability/accessibility.

Why it matters: Provides actionable, research-backed principles for designing digital learning materials that avoid cognitive overload and enhance engagement, directly applicable to HCI design and learning sciences instruction.

Method: Narrative literature review synthesizing peer-reviewed research from 2015-2024, integrating multiple theoretical frameworks including Cognitive Theory of Multimedia Learning and Self-Regulated Learning theory.

Jargon: Self-regulated learning - learners' ability to monitor, control, and reflect on their own learning process; Cognitive overload - when information exceeds working memory capacity, hindering learning.

Finding: CBT for eating disorders lacks evidence-based support for neurodivergent and Indigenous populations, with research designs creating a "methodological echo chamber" that reproduces findings from predominantly white, Western, neurotypical samples while potentially causing harm to excluded communities.

Why it matters: This challenges assumptions about universal applicability of "evidence-based" interventions and highlights how research design choices can perpetuate bias and exclusion, with direct implications for inclusive design and behavioral interventions.

Method: Draws on decolonial and neurodiversity scholarship combined with lived experience literature to critique the external validity limitations of CBT trials.

Jargon: Iatrogenic harm = harm caused by medical/therapeutic treatment itself; neuronormative = assuming neurotypical ways of thinking/processing as the standard; external validity = how well research findings generalize to other populations/contexts.

Finding: Introduces the Presence Without Belonging (PWB) framework showing that autistic loneliness often occurs during active social participation when conditions for true belonging—recognition, access, and sustainability—are missing, reframing loneliness as structural rather than individual deficit.

Why it matters: Provides crucial insights for designing inclusive team environments and collaborative technologies that support genuine belonging rather than mere participation, directly relevant to HCI accessibility and team dynamics research.

Jargon: Epistemic justice - being recognized as a credible knower and having one's social experiences validated; Neurodiversity - viewing neurological differences like autism as natural variations rather than deficits.

Finding: This study directly compared two experimental paradigms for measuring habits in humans and found that time pressure affects habit expression differently depending on the task - the Response Remapping task showed expected stronger habits under time pressure, while the Outcome Devaluation task counterintuitively showed more habitual behavior with longer preparation time.

Why it matters: Understanding how time pressure influences automatic versus deliberate decision-making is crucial for designing interfaces, team processes, and behavioral interventions that account for when people rely on habits versus conscious choice.

Method: Systematic experimental comparison of two habit measurement paradigms with manipulated response time constraints across two experiments.

Jargon: Outcome Devaluation task - measures habits by testing if people continue actions even when rewards are removed; Response Remapping task - measures habits by changing action-outcome mappings and seeing if people stick to learned responses.

Finding: A tongue-computer interface using surface electromyography achieves 88.4% accuracy for silent speech recognition through tongue movements alone, without requiring vocal cords, throat movement, or sound production.

Why it matters: This represents a novel HCI modality that could transform accessibility design and human-computer interaction for users with speech impairments or in noise-constrained environments.

Method: Used soft electrode arrays to capture tongue muscle electrical signals (sEMG) and developed a multiscale attention fusion temporal convolutional network (MSAF-TCNet) for pattern recognition across seven participants.

Jargon: sEMG = surface electromyography (measuring electrical activity of muscles through skin surface); TCI = tongue-computer interface.

Finding: Gamification experience in AR-based game tourism significantly enhances travel memory vividness and place attachment, which both mediate the relationship between game tourism motivation and revisit intention, with seven distinct tourist experience configurations identified through fuzzy-set analysis.

Why it matters: Provides empirical evidence for how gamification design elements influence user motivation, memory formation, and attachment—directly applicable to designing gamified systems and understanding the psychological mechanisms of game-based experiences.

Method: Combines covariance-based structural equation modeling with fuzzy-set qualitative comparative analysis to identify both linear relationships and complex configurations of user experience factors.

Jargon: fsQCA = fuzzy-set qualitative comparative analysis, a method for identifying different combinations of conditions that lead to outcomes; place attachment = emotional bond between person and place.

Finding: The study developed and evaluated an augmented reality application (Magical SeDu) for sexual education of children with special needs, achieving 86.04% user satisfaction through a user-centered design approach involving stakeholders throughout development.

Why it matters: This demonstrates how AR technology can create inclusive, interactive learning experiences for underserved populations, offering insights for designing accessible educational interfaces and inclusive UX practices.

Method: Used a three-phase instructional design model (planning, design, development) followed by user acceptance testing to validate usability and effectiveness.

Jargon: UAT (User Acceptance Testing) - final testing phase where actual users validate that a system meets their requirements and works as expected.

Finding: Researchers developed an AR framework for teachers based on Gagne's nine events of instruction and self-determination theory, identifying five key constructs (motivation, technology skills, instructional design, AR development tools, and AR application types) needed to enhance student motivation through AR learning experiences.

Why it matters: This provides a structured approach for designing educational technology that supports motivation and learning, directly relevant to HCI design for education and understanding how technology interfaces can be designed to support intrinsic motivation.

Method: Used design and development research methodology with mixed-methods data collection from 35 technology teachers through purposive sampling.

Jargon: Self-determination theory (SDT) - psychological framework focusing on intrinsic motivation through autonomy, competence, and relatedness; Gagne's nine events - instructional design model with nine sequential teaching steps.

Finding: This study develops a computational thinking-integrated writing module for ESL classrooms that embeds CT elements (decomposition, pattern recognition, abstraction, algorithmic thinking, and logical reasoning) into English writing instruction using visualization tools and inquiry-based approaches.

Why it matters: This demonstrates how computational thinking frameworks can be applied to enhance traditional skill acquisition (writing) and shows practical integration of CT concepts into educational design.

Method: Used ADDIE instructional design model with focus group validation from ESL experts, incorporating constructivist and experiential learning theories.

Jargon: ADDIE = Analysis, Design, Development, Implementation, Evaluation model for instructional design; CT = computational thinking involving problem-solving approaches from computer science.

Finding: Future EFL teachers in Indonesia hold predominantly positive perceptions of generative AI tools like ChatGPT and Gemini, recognizing their educational utility while expressing concerns about practical implementation in blended learning contexts.

Why it matters: This directly addresses AI's impact on education and provides empirical evidence about teacher attitudes toward AI integration, which is crucial for understanding how AI tools will be adopted in learning environments.

Method: Mixed-methods study with 150 participants using surveys and semi-structured interviews, analyzing quantitative data for frequency distributions and qualitative data through interactive model analysis.

Jargon: EFL = English as a Foreign Language teaching; critical digital pedagogy = educational approach that critically examines how digital technologies can enhance learning while addressing power dynamics and equity issues.

Finding: This paper provides a comprehensive introduction to generative AI, covering its foundations, mechanisms, applications across creative industries, education, and human-AI collaboration, along with emerging research directions.

Why it matters: Offers foundational knowledge essential for understanding AI's transformative impact on creative work, educational practices, and collaborative processes across the domains you teach.

Method: Appears to be a comprehensive review/survey paper synthesizing current knowledge in the field.

Jargon: GenAI = Generative Artificial Intelligence (AI systems that create new content); multimodal artifacts = content combining multiple formats like text, images, and audio.

Finding: This narrative review examines how AI-based technologies can identify, prevent, and manage school-related anxiety in adolescents through personalized digital interventions, affective computing, and intelligent tutoring systems. The analysis integrates psychological frameworks with inclusive education perspectives while addressing ethical concerns about privacy, bias, and equitable access.

Why it matters: This directly addresses AI's impact on learning environments and student wellbeing, offering insights for designing supportive educational technologies and understanding how AI can enhance human-centered educational experiences.

Method: Integrative narrative review of literature from 2010-2025 focusing on AI-supported mental health applications in educational settings.

Jargon: Affective computing - technology that recognizes and responds to human emotions; Universal Design for Learning (UDL) - educational framework providing multiple ways to engage, represent information, and express learning.

Finding: The paper argues that US immigration discourse operates through emotional governance rather than rational policy reasoning, using fear, disgust, and urgency to make coercive practices publicly acceptable through psychological mechanisms like dehumanization and moral disengagement.

Why it matters: This demonstrates how emotional framing and cognitive biases shape public discourse and decision-making, which is directly relevant to understanding behavioral science principles in political and social contexts.

Jargon: Authoritarian drift = gradual erosion of democratic norms through psychological rather than forceful means; Affective diffusion = spread of emotional responses through populations; Moral disengagement = psychological mechanisms that allow people to justify harmful actions.

Finding: Testing 165 human adults, researchers found that associative symmetry (the ability to reverse learned associations) - considered a uniquely human cognitive capacity - is much more fragile than previously thought, with minor protocol changes dramatically disrupting performance.

Why it matters: This challenges assumptions about fundamental human cognitive abilities that underlie learning and decision-making, relevant to understanding how people acquire skills and process information in educational and collaborative contexts.

Method: Three experiments systematically varied the standard associative symmetry protocol to test robustness, emphasizing minimal verbal instructions to reduce task-related biases.

Jargon: Associative symmetry - if you learn A leads to B, you can automatically infer B leads to A; proto-symbolic capacity - basic cognitive building blocks that support symbolic thinking.

Finding: This theoretical review proposes that moral decision-making follows a temporal cascade across three brain networks: salience network for detecting moral cues, central executive network for conflict resolution, and default mode network for value-based evaluation, each with distinct neural signatures.

Why it matters: Understanding the neural mechanisms of moral decision-making provides insights into how people navigate ethical choices in teams, organizations, and human-AI collaboration contexts.

Jargon: ERP (event-related potentials) = brain electrical responses to stimuli; P200/P260/N200/P300/LPP = specific neural response components at different time intervals; theta/delta/alpha/beta = different brain wave frequencies.

Finding: Converting tactile sensor time-series data into image representations (GAF and MTF) and processing them with CNNs achieves over 95% accuracy in contact state recognition with millisecond-level inference latency.

Why it matters: This demonstrates how novel data encoding techniques can significantly improve tactile interface performance, which is crucial for developing more intuitive and responsive human-computer interaction systems.

Method: Uses a 4×4 PVDF sensor array with time-series to image conversion (Gramian Angular Field and Markov Transition Field) processed through convolutional neural networks.

Jargon: PVDF (polyvinylidene fluoride) - a flexible piezoelectric material used in pressure sensors; GAF/MTF - mathematical transformations that convert time-series data into 2D image representations.

Finding: The paper describes an AI assistant that combines voice recognition, hand gesture recognition, and emotion detection to provide personalized, contactless human-computer interaction, including adaptive responses like playing calming music when negative emotions are detected.

Why it matters: This demonstrates multimodal interaction design and emotion-aware computing, which are important areas in HCI for creating more natural and responsive user interfaces.

Method: Uses OpenCV and DeepFace library for emotion detection, Gemini 2.5 Flash for voice interpretation, implemented as a Python desktop application with real-time system control capabilities.

Jargon: DeepFace - a deep learning library for facial emotion recognition; OpenCV - computer vision library for image processing.

Finding: Vision transformers achieve 81% accuracy in recognizing facial emotions when people wear face masks, outperforming traditional deep learning approaches by using self-attention mechanisms and synthetic mask augmentation.

Why it matters: This directly advances human-computer interaction by improving emotion recognition systems that must work in real-world conditions where facial occlusion is common, enabling better affective computing interfaces.

Method: Fine-tuned a pre-trained vision transformer on the AffectNet dataset, restructuring eight emotion categories into five broader classes and using synthetic mask augmentation to simulate occlusion.

Jargon: Vision transformers (ViT) - neural networks that use self-attention mechanisms instead of convolutions to process images; AffectNet - large-scale facial emotion dataset.

Finding: This review analyzes AI governance frameworks across the U.S., EU, and Africa, arguing that regulatory responses should be context-specific and grounded in ethical principles while addressing concerns about bias, transparency, and unequal distribution of AI benefits.

Why it matters: Provides essential background on the regulatory landscape shaping AI's impact on work and society, which is crucial for understanding constraints and opportunities in AI-human collaboration research.

Method: Comparative policy analysis across three major regulatory regions.

Finding: Pre-service early childhood education teachers in Lesotho experienced blended learning as offering flexible learning options and improved collaboration, but faced significant challenges from limited technology access and insufficient support, though their perceptions didn't correlate significantly with diffusion of innovation theory variables.

Why it matters: This demonstrates how technology adoption barriers affect learning experiences and collaborative outcomes, providing insights for designing better educational technology implementations and support systems.

Method: Mixed-methods study combining focus groups (6 participants) with surveys (271 respondents), analyzing qualitative data thematically and quantitative data through descriptive and inferential statistics using Pearson correlations.

Jargon: Diffusion of innovation theory - framework explaining how new technologies spread through populations over time based on factors like perceived advantages and complexity.

Finding: Vodcasts incorporating PhET simulations significantly improved Grade 12 STEM students' understanding of vector concepts in physics, achieving a normalized gain of 0.68 and receiving satisfactory ratings from evaluators.

Why it matters: Demonstrates how multimedia educational tools can enhance learning of complex STEM concepts, relevant for understanding effective instructional design and technology-enhanced learning.

Method: Used ADDIE (Analysis, Design, Development, Implementation, Evaluation) model to develop vodcasts and measured learning gains through pre/post assessment.

Jargon: PhET simulations - interactive physics simulations from University of Colorado; vodcasts - video podcasts combining audio and visual elements; normalized gain - standardized measure of learning improvement.

Finding: A phenomenological study of 62 stakeholders found that school gardening programs enhance collaborative learning, experiential skill acquisition, and behavioral development in elementary students, but face implementation barriers from institutional constraints and stakeholder resistance.

Why it matters: Provides empirical evidence on how hands-on experiential learning environments can foster collaboration and skill acquisition, relevant to understanding alternative learning approaches and team dynamics in educational settings.

Method: Qualitative phenomenological study using Colaizzi's descriptive method to analyze semi-structured interviews with diverse stakeholders (students, teachers, parents, administrators).

Jargon: Phenomenological study - research approach focused on participants' lived experiences and subjective meanings; Colaizzi's method - specific framework for analyzing phenomenological data through systematic steps.

Finding: Service learning in doctoral public health education enhances both student learning outcomes and community engagement through participatory action research and capacity development, analyzed through social learning theory, theory of planned behavior, and self-determination theory.

Why it matters: This demonstrates how structured experiential learning can simultaneously achieve educational goals and community impact, relevant to understanding motivation, skill acquisition, and collaborative engagement in educational settings.

Method: Four-year longitudinal study with qualitative document analysis of reflective reports from 76 DrPH students implementing community health interventions.

Jargon: Service learning - educational approach combining academic learning with community service; DrPH - Doctor of Public Health degree.

Finding: An Arduino-based Geiger-Muller counter significantly improved university physics students' critical thinking skills (evaluation, analysis, interpretation, inference) when used as a hands-on learning tool for radioactivity concepts.

Why it matters: Demonstrates how interactive technology tools can enhance critical thinking and skill acquisition in STEM education, relevant to learning sciences and educational technology design.

Method: Pre-experimental design with 165 students using pretest-posttest measurement and ADDIE development model for the learning tool.

Jargon: ADDIE = Analysis, Design, Development, Implementation, Evaluation instructional design framework; GMC = Geiger-Muller Counter for detecting radiation.

Finding: This paper presents Regression Discontinuity (RD) design as a quasi-experimental method for evaluating classroom interventions in STEM education when randomized trials aren't feasible, comparing students just above and below intervention thresholds to estimate causal effects.

Why it matters: This provides education researchers with a rigorous methodological approach to evaluate learning interventions and skill acquisition programs when traditional experimental designs aren't practical.

Method: Uses real classroom data to demonstrate RD analysis with step-by-step R code, exploiting predetermined intervention thresholds to compare students who narrowly qualified versus those who narrowly missed qualification.

Jargon: Regression Discontinuity (RD) design - a quasi-experimental method that estimates treatment effects by comparing outcomes of individuals just above and below an arbitrary threshold that determines treatment assignment.

Finding: This paper proposes an "AI Empowerment Hierarchical Practice Collaborative Innovation" framework for robot education that integrates theory and practice through a "Perception-Integration-Control" curriculum system and "Learning-Practicing-Innovating" teaching progression.

Why it matters: This demonstrates how AI can transform educational approaches by supporting personalized learning paths and human-machine collaborative teaching models, relevant to AI's impact on learning sciences and skill acquisition.

Jargon: Virtual-real integrated practice platform - combining physical robotics with virtual simulation environments for hands-on learning.

Finding: This bibliometric analysis of 856 papers reveals that AI applications in textile and fashion design are rapidly evolving, with research peaking in 2024 and shifting toward electronic textiles, wearable health technologies, and large language models.

Why it matters: Shows how AI is transforming creative design processes in fashion, offering insights into human-AI collaboration in creative industries and the broader impact of AI on design work.

Method: Bibliometric analysis using Web of Science database and VOSviewer software to analyze publication trends, citations, and keyword evolution from 1980-2024.

Jargon: Bibliometric analysis - statistical analysis of publications to understand research trends and influence patterns in a field.

Finding: The paper proposes a multi-scale Shapley adaptation pruning method to defend brain-computer interface systems against backdoor attacks by identifying and removing malicious weights in neural networks while maintaining EEG classification accuracy.

Why it matters: This addresses security concerns in human-computer interaction systems that rely on neural signals, which is relevant to HCI but focuses heavily on technical security rather than user experience or interaction design.

Method: Uses Shapley values to locate backdoor weights in neural networks and applies adaptive pruning with a cost function, tested on BCI competition datasets.

Jargon: BCI (brain-computer interface) - systems that enable direct communication between brain and computer; backdoor attack - malicious modification of AI models to trigger incorrect behavior; Shapley values - method for attributing contributions of individual features in machine learning models.

Finding: Principal leadership and academic supervision significantly influence learning quality and teacher performance in high schools, with teacher performance being the dominant factor (72.1%) in improving learning quality, though work motivation does not moderate this relationship.

Why it matters: This provides empirical evidence about leadership effectiveness and motivation factors in educational settings, relevant to understanding team dynamics and behavioral factors in organizational performance.

Method: Used quantitative path analysis and Moderated Regression Analysis (MRA) with proportional random sampling of teachers across public high schools.

Jargon: MRA (Moderated Regression Analysis) - statistical technique to test whether a third variable affects the relationship between predictor and outcome variables.

Finding: This paper examines how AI systems represent and structure knowledge using formal systems like logic, graphs, and ontologies to enable machine reasoning and problem-solving.

Why it matters: Understanding knowledge representation is relevant for designing AI systems that collaborate effectively with humans and for comprehending how AI tools impact cognitive work processes.

Jargon: Knowledge Representation (KR) - the field focused on encoding human knowledge in formats that computers can process; ontologies - formal specifications of concepts and their relationships in a domain.

Finding: This study examines how high school music achievement standards can be aligned with digital literacy development using Bloom's Digital Taxonomy, focusing on collaborative learning and communication tools in digital music classrooms.

Why it matters: The focus on digital literacy development through collaborative learning in educational settings connects to learning sciences and team dynamics research.

Method: Analyzed music achievement standards and consulted three digital education experts to assess alignment with digital literacy development using Bloom's Digital Taxonomy framework.

Jargon: Bloom's Digital Taxonomy - an adaptation of Bloom's educational taxonomy specifically focused on digital learning objectives and skills.