Middle-income states like Thailand face a structural dilemma: EU-style AI regulation exceeds administrative capacity, while voluntary models fail to protect fundamental rights. Leveraging Thailand’s 2025 BRICS Partner status, this study proposes a Thai–BRICS Hybrid Governance Model based on functional modularity. This approach avoids wholesale transplantation, instead selectively adapting regulatory mechanisms from BRICS nations to fit Thailand’s specific legal and fiscal constraints. The model addresses five critical gaps: infrastructure dependency, algorithmic opacity, accountability deficits, institutional fragmentation, and labor displacement. The study’s central thesis is that rights remain symbolic without developmental sovereignty, the material control over digital infrastructure. By prioritizing sovereign capacity, Thailand can ensure that algorithmic accountability is enforceable rather than aspirational. This framework reconciles human rights with developmental goals, avoiding the prohibitive compliance burdens seen in previous GDPR-inspired legislation and positioning infrastructure as a prerequisite for genuine rights protection.

This study examines the attitude formation process in viewers of AI-based fashion films by focusing on cognitive, affective, and perceptual responses. As generative AI reshapes the production and visual language of fashion media, fashion films are no longer limited to the documentation of physical garments but function as visual media that construct aesthetic experience. Within this context, the study explores how viewers interpret and evaluate AI-generated fashion imagery, with particular attention to the roles of meaning, emotion, and visual perception in shaping attitude. An empirical approach was employed using AI-generated fashion film content as a stimulus. Participants evaluated their cognitive, affective, and perceptual responses to design elements presented in the film. The findings indicate that perceptual factors are statistically significantly associated with attitude, whereas cognitive and affective factors are not. Visual elements such as silhouette, color, material, and detail serve as key cues that facilitate immediate and intuitive judgment. These results suggest that the attitudes of AI-based fashion film viewers are significantly shaped by perceptual judgments of visual elements. The study contributes to the understanding of AI-based fashion films as a form of fashion communication and offers implications for the development of fashion content that emphasizes perceptual clarity, visual coherence, and the effective articulation of design elements.

This study explores creative experiences and educational needs related to the use of generative artificial intelligence (AI) in fashion design and proposes educational strategies through an AI-based fashion design process framework. A qualitative research design was employed, involving semi-structured in-depth interviews with 10 fashion design educators who had experience using generative AI. Inductive content analysis was performed on the collected data using NVivo, comprising coding, categorization, and theme development. The findings were organized into four major themes: (a) perceptions of AI use in fashion design, (b) functional roles of AI in the creative process, (c) human–AI collaboration and creative agency, and (d) educational needs and ethical considerations. The results showed that generative AI was perceived not as a substitute for human designers but as a supportive tool that could enhance creative thinking, particularly in the ideation and visualization stages. Specifically, AI enabled rapid exploration of diverse design alternatives and reduced psychological pressure in early creative phases. Human–AI collaboration was characterized by a complementary structure in which AI generated visual suggestions but human designers retained the responsibility for aesthetic judgment and final decision-making. Finally, an AI-based fashion design process framework aligned with the Double Diamond model was derived from these findings, providing a conceptual basis for educational strategies in fashion design education.

본 연구는 고령층의 사회보장 정보 접근성 제고를 위해 AI 기반 복지상담 서비스의 수용 요인을 분석하고, 디지털 정보격차 완화를 위한 정책적 시사점 을 도출하고자 하였다. 분석 자료는 한국지능정보사회진흥원(NIA)이 실시한 20 23년 「디지털 정보격차 실태조사」에 응답한 만 65세 이상 고령자를 대상으로 하였으며, 다중회귀분석을 실시하였다. 분석 결과, 디지털 접근, 디지털 역량, 디지털 활용 지수는 모두 AI 기반 복지상담 서비스 수용도와 유의한 정(+)의 관련성을 보였다. 또한 기초연금 인지도와 노인장기요양서비스 이용 경험은 디 지털 요인을 통제한 이후에도 AI 상담 서비스 수용도와 유의미한 관련성을 나 타냈다. 이는 기존 복지제도에 대한 인식과 이용 경험이 신기술 기반 복지서비 스 수용과 연관될 수 있음을 시사한다.

본고는 빅데이터 시대의 새로운 연구 방법론으로 부상한 AI 기반 토픽모델링 (BERTopic)을 고고학 연구동향 분석에 적용하고, 그 실체적 효용성과 한계를 규명하는 데 목적이 있다. 이를 위해 2006년부터 2013년까지 한국 청동기시대 석기 관련 연구 논 문 75편을 대상으로 텍스트 마이닝을 수행하였다. 기존의 빅데이터 기반 연구동향 분석이 거시적 경향성 파악에 치중하여 개별 연구의 논리적 맥락을 소거하는 한계를 극복하고자, 본고는 전문가의 정성적 검토가 가능한 규모로 데이터셋을 통제(Controlled Dataset)하여 AI 분석 결과의 미시적 정합성을 정밀하게 검증하였다. 이후 그 결과를 동일시기의 연구 성과를 정량·정성적으로 고찰한 연구사적 논문 결과(손준호 2013)와 직접 비교·분석을 진 행하였다. 텍스트 분석 결과, AI는 방대한 문헌 속에서 ‘형식·편년 중심’과 ‘생산·생계 중심’이라는 거시적 연구 지형을 신속하게 파악하고, 텍스트 이면에 잠재된 방법론적 맥락(자연과학적 분석 등)을 수치로 입증하는 데 탁월한 효용을 보였다. 그러나 미시적 분석 단계에서는 비 판적 논조를 파악하지 못하는 ‘문맥 소거’, 이질적인 시공간 데이터를 기계적으로 결합하 는 ‘사실 왜곡’, 그리고 연구의 질적 경중을 가리지 못하는 ‘가치 평가 부재’라는 결정적 한계를 드러냈다. 이에 필자는 AI의 연산 능력을 맹신하는 태도를 경계하고, 연구자의 경험적 통찰이 AI 의 기계적 객관성을 보완하는 ‘전문가 매개(Expert-Mediated) 통합 분석 모델’을 제안하 였다. 이는 AI에게 1차적인 데이터 처리와 지도 작성을 맡기되, 사실 검증(Fact Verification), 논리적 맥락의 복원(Contextual Calibration), 연구사적 가치 부여 (Qualitative Valuation)의 최종 권한은 인간 연구자가 수행해야 함을 의미한다. 결론적 으로 디지털 고고학의 미래는 데이터의 양적 팽창에 함몰되지 않고, 연구자의 전문적 식견 을 통해 데이터에 학술적 생명력을 불어넣는 방향으로 나아가야 함을 역설하였다.

Injection-molded products frequently exhibit localized surface defects such as weld lines, flow marks, scratches, bubbles, and burn marks due to variations in material flow, mold temperature, and cooling conditions. Conventional visual inspection is highly dependent on operator experience, while rule-based machine vision methods are limited under variations in lighting and surface texture. This study proposes a deep learning–based defect detection model using YOLOv8 combined with a novel Defect-Aware Augmentation technique designed to enhance robustness for small, local defect regions. The proposed augmentation pipeline includes geometric transformations, optical perturbations, local defect patch synthesis, and diffusion-based synthetic defect generation. Experiments were conducted on a custom dataset of 5,000 images (3,000 normal and 2,000 defective). Results show that the proposed model achieves significant improvements over baseline models, obtaining 95% precision, 90% recall, and 0.96 mAP@0.5, outperforming the default YOLOv8 model by 7%p in mAP. Ablation studies verify that defect-aware augmentation is the dominant factor contributing to the performance gain. The proposed system demonstrates high applicability for automated quality inspection in injectionmolding production lines.

Background: Stroke often leads to persistent gait impairments that significantly reduce mobility and quality of life. Conventional rehabilitation has demonstrated therapeutic value but is limited by insufficient personalization and low patient engagement. Objectives: This study aimed to evaluate the clinical effectiveness of a realtime Kinect-based motion analysis and AI-driven virtual reality (VR) gait training system for stroke rehabilitation. Design: Randomized controlled trial with parallel-group assignment. Methods: Thirty stroke patients were randomly assigned to a VR-based gait training group (n=15) or a conventional physical therapy group (n=15) for 8 weeks. The VR system integrated Kinect-based markerless motion capture, a 14-layer artificial neural network for gait parameter prediction, and immersive VR feedback to provide personalized gait retraining. Spatiotemporal gait parameters—including gait velocity, step length, cadence, and step width— were assessed before and after the intervention. Results: The VR group demonstrated significantly greater improvements in gait velocity (0.52 to 0.73 m/s, +40.4%), step length (78.3 to 95.7 cm, +22.2%), and cadence (100.2 to 110.4 steps/min, +10.2%) than the control group, while step width decreased (12.3 to 9.8 cm, −20.3%), indicating enhanced balance and stability. The artificial neural network accurately predicted movement patterns and supported adaptive training with real-time feedback. Conclusion: The real-time VR gait rehabilitation system effectively enhanced gait performance and motor coordination among stroke patients, outperforming conventional physical therapy. The integration of Kinect-based motion capture and AI-driven personalization provides a promising platform for scalable and clinically meaningful stroke rehabilitation.

Despite over two decades of commercialization, conversational AI continues to produce functional and communicative errors. This study examines how users’ cultural and linguistic backgrounds influence their experience with conversational AI as well as their error tolerance and adoption rates. We hypothesize that users’ cultural and linguistic backgrounds affect both error tolerance and user experience and that the number of languages a user speaks may amplify this effect. To evaluate this, we conducted in-depth qualitative, 1-hour interviews with eight multilingual users based on a standardized set of questions. All sessions were conducted online on Zoom and were audio-recorded. Results revealed that multilingual users face more diverse linguistic and cultural challenges yet demonstrate greater error tolerance, often continuing to use AI for basic tasks despite inconveniences. Furthermore, monocultural and language users are more likely to discontinue use when errors persisted. By adapting to diverse users, conversational AI can enhance user experience, reduce disparities, and promote equitable access. This study provides insights for developing inclusive, sustainable, and socially responsible conversational AI systems accessible to a global user base. However, limitations include the narrow diversity of participants’ countries and languages and a small sample size. Future research should expand participant diversity to provide a more comprehensive and deeper understanding of conversational AI systems.

본 연구는 ADHD 아동의 정서 조절 어려움과 부모-자녀 의사소통 단절 문제를 완화하기 위해, 생성형 AI를 활용한 관계 촉진형 워크북 시스템의 설계 프레임워크를 제안하는 설계 연구(Design Research)로서, 이론적으로 근거한 설계 원칙의 도출과 기술적 실현 가능성 검 증에 초점을 두며, 사용성 평가 및 임상적 효능 검증은 후속 연구 과제로 설정한다. 제안 시스템은 아동의 일기를 4컷 만화로 변환하되 말풍선을 비워 둠으로써 부모-자녀가 대화를 통해 의미를 공동 완성하도록 하는 '의도적 불완전성' 설계 패러다임을 적용하여, AI가 치 료를 대체하지 않고 상호작용의 비계로 기능하도록 한다. Barkley의 실행 기능 결함 모델, Fogel의 공동 조절 이론, PCIT의 PRIDE 기술, Sweller의 인지 부하 이론이라는 4가지 이론 적 출처의 교차 분석을 통해 5가지 설계 원칙을 도출하고, LLM 및 Text-to-Image 모델 기반 프로토타입을 구현하여 기술적 실현 가능성을 검증하였으며, 전문가 3인의 의견을 수렴하 여 설계의 현장 적합성을 탐색하였다. 본 연구는 일상적 일기 쓰기를 정서 교류 기회로 전 환할 가능성을 논의하며, 향후 파일럿 사용성 연구 및 임상 검증을 위한 기반을 제공한다.

노후화된 사회 기반 시설물 증가에 따라 정기적인 구조물 손상 점검의 중요성이 확대되고 있다. 그러나 기존 점검 방식은 고가의 장비와 다수의 인력을 요구하며, 차선 폐쇄를 필수적으로 수반한다. 특히 차선 폐쇄는 교통 체증을 유발해 차량의 반복적인 가속과 감속, 공회전을 증가시키고 결과적으로 연료 소비와 온실가스 배출량을 증가시켜 사회적 비용을 초래한다. 이에 AI 기술을 활용해 차선 폐쇄 없이 손상을 탐지하는 연구가 진행되고 있으나 대부분 도로포장 탐지에 한정되어 있어 교량 기둥이나 방호 울타리 등 입체 구조물에 대한 탐지 기술과 차선 폐쇄에 따른 운영 효율성 및 에너지와 배출량 변화에 대한 정량적 분석은 부족한 실정이다. 본 연구는 차선 폐쇄 없이 사회 기반 시설물의 손상을 탐지할 수 있는 AI 기반 손상 시스템을 구축하고 차선 폐쇄로 인한 변화를 정량적으로 분석한다. 이를 위하여 360° 카메라, 차량 전방 카메라, 라인 스캔 카메라를 통하여 도로 영상을 수집하고, Mask R-CNN과 RF DETR+SAM 알고리즘을 활용하여 도로포장과 입체 구조물의 손상을 탐지하였다. 또한, 교통 시뮬레이션 프로그램 SUMO를 통해 국내 도로 구간을 재현하고 차량 에너지 분석 모듈 FASTSim을 연계하여 차선 폐쇄에 따른 교통 및 에너지 효율 변화를 비교하였다. AI 탐지 결과 RF DETR+SAM 시스템은 정확도 81%, 정밀도 87%, 재현율 61%, F1-score 0.72를 달성해 Mask R-CNN 대비 우수한 성능을 기록했으며, 도로포장뿐만 아니라 입체 구조물에 대한 안정적 탐지 가능성을 확인하였다. 시뮬레이션 결과 차선 폐쇄는 주행 속도 약 25% 감소, 연료 소모 약 18% 증가, CO2 배출량이 약 22% 증가한 것으로 나타났다. 본 연구는 AI 기반의 손상 탐지가 차량흐름을 유지하며 수행될 수 있음을 실증하고, 유지관리 시 교통, 에너지, 환경 영향을 통합적으로 고려할 수 있는 정량적 근거를 제시한다.

This study explores the use of Midjourney (V6) by fashion design undergraduates for AI-supported ideation, focusing on how outcomes differ based on fashion-domain competence and prompt/parameter instruction. A focused ethnographic, comparative case-study design was used to observe a short collection-development module. Data included Discord prompt and parameter logs, generated image outputs (mood boards, look proposals, and pattern drafts), one-on-one interviews, classroom observation notes, and expert co-coding and qualitative evaluation. Participants were organized into four groups by crossing Basic vs. Advanced Fashion competence (BF/AF) with Basic vs. Advanced Prompt training (BP/AP): BF-BP, AF-BP, BF-AP, and AF-AP. BF-BP depended on repetitive/imaginary use and generic descriptors, resulting in visually appealing yet conceptually fragmented and low-feasibility results. AF-BP leveraged a richer domain vocabulary to improve item-level adequacy but struggled to maintain collection-level consistency, leading to the use of external editing tools such as Photoshop and Illustrator for portfolio-level refinement. BF-AP quickly mastered commands and parameters (e.g., /describe, --chaos, --stylize, --ar, --tile, --no, --sref, --cref), generating appealing concept imagery while failing to convert outputs into wearable garments and cohesive collections. AF-AP combined advanced fashion knowledge with strategic parameter sequencing— broad exploration, followed by consistency control and selective refinement—achieving the most coherent, feasible outcomes and positioning AI as an early-stage accelerator rather than a substitute for core design and making skills. Overall, this study proposes “parameter literacy” as a domain-specific extension of GenAI literacy and offers a parameter–process mapping (divergent generation, consistency control, and editing/refinement) to enhance fashion curricula.

This study examines group differences in AI dependence, self-efficacy, and design fixation based on learners’ AI utilization experience, and further explores the relationships between AI dependence and perception-related variables within an AI-based design education context. To this end, I surveyed 42 learners who participated in an AI-based design class. I then performed data analysis in IBM SPSS Statistics 26, using one-way analysis of variance (ANOVA) and Pearson correlation. The ANOVA revealed statistically significant group differences in AI dependence by AI usage level, but no significant differences in self-efficacy or design fixation. Furthermore, age and professional experience showed no significant influence on most variables. Meanwhile, the correlation analysis revealed that AI dependence was significantly positively associated with design fixation, but not with self-efficacy. These results suggest that, in AI-based design education, learners’ perceptions and attitudes are not strongly differentiated by personal background factors such as age or professional experience, but are instead associated with their experience with AI use and perceived dependence on AI. By distinguishing and analyzing AI utilization and AI dependence, this study provides empirical evidence that contributes to a more nuanced understanding of learners’ cognitive perceptions in AI-based design education in practice.

This paper presents an AI-based PHM (Prognostics and Health Management) framework for quantitative motor health assessment and remaining useful life (RUL) prediction. The proposed method first defines a health index using vibration and current signals of an industrial motor, and then adopts a two-stage PHM architecture consisting of health-state classification and deep learning-based RUL prediction. A degradation test bench is designed to obtain condition monitoring data for normal, warning, and critical states, and a hybrid 1D CNN– BiLSTM–attention model is developed to capture both local features and long-term temporal dependencies. Experimental results demonstrate that the proposed model outperforms conventional SVM and single LSTM baselines in terms of both health-state classification accuracy and RUL prediction accuracy, achieving a 20– 30% reduction in RMSE and more than 80% of RUL predictions within ±10% error. The proposed approach provides a practical PHM framework and modeling guidelines for implementing condition-based maintenance of electric motors in smart manufacturing environments.