Artificial intelligence-driven paradigm shifts in functional food ingredient research

Malnutrition and chronic diseases have increased the need for functional food ingredients (FFI) from natural sources, yet FFI discovery remains slow and fragmented. More than 93% of known food-derived bioactive molecules are still functionally unvalidated, reflecting a persistent gap between chemical diversity and biological confirmation. This review examines how FFI research is evolving across three interconnected paradigms: experience-guided experimentation, physics-based computation, and artificial intelligence (AI)-driven discovery. Traditional approaches remain indispensable for mechanistic rigor and experimental validation, but they are often labour-intensive, low-throughput, and costly. Physics-based tools, including quantum mechanics, molecular docking, and molecular dynamics simulation, improve efficiency by enabling rational solvent selection, virtual screening, and mechanism elucidation. AI-driven approaches further expand the field by supporting high-dimensional prediction, active learning, and generative design. Rather than replacing earlier paradigms, the emerging direction is their integration into a closed-loop hybrid framework in which experiments provide ground truth, computational models provide mechanistic interpretation, and AI accelerates candidate prioritization and design. We further discuss the need for quantifiable performance metrics, explicit documentation of failure modes, high-quality standardized datasets, interpretable models, and trustworthy AI governance, including a five-layer defense stack to reduce hallucination risk in generative models. Finally, we highlight self-driving laboratories and multi-model integration as future routes toward scalable, verifiable, and intelligent FFI discovery across the vast chemical and biological space of natural sources.