Neural Computing and Applications Software

Trustworthy generative artificial intelligence and large language models for Big Data: databases, mining, and ethics at scale

Dritsas, Elias, Trigka, Maria
Springer Nature · 2026-11-12 · DOI 10.1007/s00521-026-12561-7

Abstract

Generative artificial intelligence (AI) is increasingly integrated into Big Data infrastructures, shifting data-intensive systems from purely extractive analytics towards synthesis-oriented workloads. This survey presents a systems-oriented synthesis of trustworthy generative AI by examining large language models, diffusion models, and probabilistic generators as computational operators that interact with data distributions, retrieval mechanisms, program synthesis, storage engines, query processors, and orchestration layers. The analysis was organized around three complementary dimensions. The first examines the formal objectives and representative computational characteristics of the generative models. The second focuses on architectural mechanisms in distributed data systems, including provenance-aware storage, governance-aware orchestration, federated and confidential execution, and monitored inference. The third addresses system-level trust requirements encompassing fairness, privacy, interpretability, accountability, and provenance. The survey is grounded in peer-reviewed literature identified through structured search queries that explicitly connect generative modeling with Big Data system architectures, using concepts such as lineage, retrieval-augmented generation, and governance as the organizing principles. It synthesizes current evidence on how generative AI influences indexing, query optimization, data mining, retrieval, and knowledge graph reasoning, under computational and regulatory constraints. It further examines evaluation from a multi-objective perspective that jointly considers predictive performance, robustness, explanation fidelity, provenance coverage, privacy budgets, energy consumption, and governance overhead, highlighting the important limitations of existing efficiency-centric benchmarks. Finally, emerging research directions are discussed, including provenance-aware generation, governance-aware deployment, and sustainability-aware scaling, while distinguishing established capabilities from active research challenges in trustworthy generative Big Data systems.

Read at publisher →