New-Generation Artificial Intelligence: Development and Challenges

Since its origins at the 1956 Dartmouth Conference, Artificial Intelligence (AI) has become a core engine of technological revolution, now manifesting across five key paradigms: big data, cross-media, collective, hybrid-augmented, and autonomous intelligence. While its integration with other sciences is fundamentally reshaping research, AI’s progress is hindered by critical challenges, ranging from technical limitations like poor interpretability and architectural bottlenecks, to conceptual flaws such as model “hallucinations” and weak reasoning, as well as significant ethical risks. Addressing these hurdles requires a paradigm shift toward the synergistic integration of data and knowledge and the development of novel computing architectures. We argue the path forward lies in a “data-driven, knowledge-guided, and physics-constrained” framework, representing a potential fifth paradigm for scientific discovery. This approach, when combined with agile governance, provides an essential roadmap for guiding AI’s responsible evolution for the benefit of all humanity.