Integrasi Embedding Multiformat untuk Representasi Semantik Big Data Smart City: Analisis Etika Penelitian, AI Ethics, dan Tantangan Publikasi Ilmiah di Era Teknologi Lanjut

Penelitian ini bertujuan menganalisis manfaat representasi semantik dan memetakan risiko etika pada pipeline embedding multimodal dalam pengolahan Big Data Smart City. Penelitian menggunakan desain sintesis literatur yang dipadukan dengan studi kasus konseptual, analisis hermeneutik teknologi, dan pemetaan risiko berbasis pipeline. Analisis mencakup lima komponen utama, yaitu Text Encoder, Visual Encoder, Cross-Modal Alignment, Fusion Layer, dan Semantic Output Layer. Hasil penelitian menunjukkan bahwa integrasi data tekstual dan visual dapat memperkaya konteks semantik, memperkuat hubungan informasi antarmodalitas, serta mempertahankan konsistensi representasi pada dokumen kebijakan yang kompleks. Namun, manfaat tersebut disertai lima risiko utama, yaitu bias representasional, privasi dan indirect disclosure, dual-use, surveillance dan profiling otomatis, serta asimetri kekuasaan informasi. Fusion Layer teridentifikasi sebagai komponen dengan risiko paling tinggi karena menggabungkan bias dan potensi penyalahgunaan dari beberapa modalitas, sedangkan Cross-Modal Alignment menunjukkan mekanisme mitigasi yang masih terbatas. Integrasi prinsip autonomy, beneficence, justice, dan consent dengan fairness, accountability, transparency, explainability, serta manajemen risiko menghasilkan kerangka evaluasi etika yang dapat diterapkan sepanjang siklus hidup sistem. Penelitian merekomendasikan penerapan ethical checkpoints, audit bias, dokumentasi dataset dan model, logging, pembatasan tujuan penggunaan, serta human oversight. Karena berbasis studi kasus konseptual, kerangka ini masih memerlukan validasi ahli dan pengujian empiris menggunakan dataset Smart City aktual.

Ethical Risk Mapping in Multimodal Embedding Pipelines for Semantic Representation of Smart City Big Data: A Literature Synthesis and Conceptual Case Study

This study aims to analyze the benefits of semantic representation and map ethical risks within multimodal embedding pipelines used to process Smart City Big Data. The study employed a literature synthesis design combined with a conceptual case study, technological hermeneutic analysis, and pipeline-based risk mapping. The analysis covered five main components: the Text Encoder, Visual Encoder, Cross-Modal Alignment, Fusion Layer, and Semantic Output Layer. The findings indicate that integrating textual and visual data can enrich semantic context, strengthen cross-modal information relationships, and maintain representational consistency in complex policy documents. However, these benefits are accompanied by five major risks: representational bias, privacy and indirect disclosure, dual use, automated surveillance and profiling, and information-power asymmetry. The Fusion Layer was identified as the component with the highest risk because it combines bias and the potential misuse of information from multiple modalities, whereas Cross-Modal Alignment still has limited mitigation mechanisms. Integrating the principles of autonomy, beneficence, justice, and consent with fairness, accountability, transparency, explainability, and risk management produced an ethical evaluation framework that can be applied throughout the system lifecycle. The study recommends implementing ethical checkpoints, bias audits, dataset and model documentation, logging, purpose limitation, and human oversight. Because the study is based on a conceptual case scenario, the proposed framework still requires expert validation and empirical testing using actual Smart City datasets.