Reliability, Retrieval, and Privacy in Database-Centered AI: A Review of Emerging Foundations for Computational Discovery

Abstract

DATAMIND's first publication year established a distinctive agenda for database-centered artificial intelligence. This review synthesizes all articles published in the journal during 2023 and connects them with eighty DOI-bearing references on continual learning, medical image analysis, retrieval-augmented generation, privacy-preserving learning, robustness, and data documentation. A structured coding design is used to classify each DATAMIND article by problem domain, data object, methodological family, evaluation focus, and governance implication. The analysis shows that the 2023 corpus can be read as a coherent movement from model performance toward evidence discipline. Transformer forgetting foregrounds the problem of memory over time; medical imaging emphasizes external validation and domain-specific data; data engineering reveals hidden infrastructure behind model quality; and GraphRAG versus VectorRAG clarifies how retrieval architecture shapes enterprise knowledge generation. The review adds two grayscale figures and three tables that summarize same-year DATAMIND evidence, the review rubric, and a research agenda. The main conclusion is that database-centered AI should be evaluated through an evidence chain that includes data provenance, retrieval design, privacy safeguards, model adaptation, and human review. This synthesis positions DATAMIND as a venue for computational discovery research in which databases are active determinants of reliability, not passive storage layers.

Keywords: Database-centered AI; retrieval-augmented generation; continual learning; medical AI; privacy; data engineering; computational discovery