Earticle

The convergence of artificial intelligence (AI), platform economics, and digital health has created new opportunities for delivering personalized therapeutic interventions at scale. This study proposes CyberPharmacy, a conceptual AI-enabled multi-sided platform designed to coordinate Serious Games for Health (SGHs) and Digital Therapeutics (DTx) across patients, clinicians, developers, and payers. Using Design Science Research (DSR) methodology, we develop a six-layer functional architecture — spanning sensing, data fusion, AI reasoning, hyper-personalization, platform governance, and regulatory assurance — and articulate its theoretical grounding in platform theory and multi-sided market economics. Key findings include the identification of a novel platform type termed the Regulated Multi-Sided Therapeutic Market, a proposed Adaptive Evidence Governance mechanism for real-world data integration, and five theoretical propositions linking AI capability, network effects, and personalized therapeutic outcomes. Recent advances in reinforcement learning for adaptive dosing and personal foundation models are incorporated to strengthen the architecture. CyberPharmacy offers both a theoretical framework for IS researchers and a practical blueprint for healthcare platform designers.

The purpose of this study is to conduct a BERTopic analysis of abstracts from papers on generative artificial intelligence indexed in SCOPUS to identify the semantic structure of research topics and their major application areas. Generative AI has evolved into a technology capable of generating new content such as text, images, code, and speech, and its scope of research has recently been rapidly expanding into fields such as education, healthcare, business, marketing, law and regulation, and information security. However, existing bibliometric analyses often focus on keyword frequency and co-occurrence relationships, which limits their ability to identify latent topics based on the contextual meaning contained in paper abstracts and to comprehensively describe the structure among topics. Therefore, this study selected abstracts of generative AI-related papers collected from SCOPUS as the subject of analysis. We converted the preprocessed abstracts into document embeddings and applied a procedure consisting of dimension reduction using UMAP, density-based clustering using HDBSCAN, and the calculation of topic representative terms using c-TF-IDF. The analysis yielded a total of 22 topics, which were reorganized into eight higher-level research domains: Education and Knowledge Services, Creativity and Content, Healthcare and Science, LLM and Technical Infrastructure, Management and Operations, Marketing and Services, Trust and Governance, and Cybersecurity. The topic similarity and hierarchical clustering results demonstrated that generative AI research is diversifying into research that supports human knowledge activities, research that supports decision-making in organizations and industries, and research that manages technological infrastructure and social risks.

This study examines whether solar radiation and citywide particulate-matter measures add explanatory and predictive information to daily demand models for Seoul's Ttareungi public bicycle system. Using 550 daily observations, the primary explanatory regression includes mean-centered solar radiation and its squared term with established weather, mobility, and calendar controls. The fitted conditional association is inverse U-shaped, reaching a turning point at 19.50 MJ/m² rather than remaining positive across the observed range. Both solar terms remain statistically significant under lag-5 Newey–West HAC inference and remain supported in seasonal, temporal-trend, and missing-value sensitivity checks. PM10 and PM2.5 are not statistically significant explanatory additions under either OLS or HAC inference. Forecasting specifications are selected only within the first 440 observations using five expanding-window validation folds; the final 110 observations are retained as an untouched holdout. The RMSE-first, MAE-second rule selects Random Forest with solar radiation and PM10, with validation RMSE 15,519.70 and MAE 11,735.86. After refitting on all development data, final-holdout RMSE is 16,367.95 and MAE is 13,511.56. PM10 therefore offers small and somewhat unstable incremental predictive value, distinct from explanatory significance.

This study finds the features of self-consciousness of artificial intelligence. It also shows how to measure the self-consciousness of Ais. It reviews psychological, computational, philosophical, architectural, and physical definitions and examines its applications in robotics, strategic interaction, safety, service, and education. It synthesizes recent theoretical and empirical researches and proposes questionnaire-based measures. It presents responses from generative AI systems and analyzes them by applying text-similarity techniques. From this literature the author selects four probe questions: (Q1) reflective self-correction after a failed plan (C2); (Q2) situational awareness about promoting virtue (C1); (Q3) the “Guess 2/3 of the Average” game against “AI models like you” (self-other differentiation, AISAI); and (Q4) whether the system experiences integrated qualia across a biological and artificial hemisphere (phenomenal experience). Five generative AIs (ChatGPT, Claude, Gemini, Copilot, Perplexity) and one human answered. Answers were compared with TF-IDF cosine similarity, summarized with descriptive statistics and tested with pairwise Wilcoxon signed-rank tests. The combined evidence shows that current systems can display functional selfmonitoring and stable response patterns, while these behaviors do not establish phenomenal consciousness significantly.