Sixth-generation (6G) wireless networks are anticipated to deliver capabilities that fundamentally exceed those of 5G, including sub-millisecond end-to-end latency, terabit-per-second peak throughput, and native support for artificial intelligence (AI) as a core network function rather than an overlay feature. Network slicing, the practice of partitioning a shared physical infrastructure into multiple logically independent virtual networks, is widely recognized as the enabling mechanism through which 6G will simultaneously serve radically heterogeneous application classes. This paper provides a systematic descriptive analysis of 6G network slicing architectures and the AI-driven resource allocation frameworks proposed to manage them, with particular emphasis on ultra-low latency application requirements. Drawing exclusively on peer-reviewed literature published between 2019 and 2026, the study characterizes the architectural principles of 6G slicing, the taxonomy of AI techniques applied to slice resource management, the latency decomposition of candidate 6G use cases, and the open challenges that define the current state of the field. The analysis identifies deep reinforcement learning, federated learning, and graph neural networks as the dominant AI paradigms in documented slicing frameworks, each addressing distinct operational dimensions of the resource allocation problem. The paper concludes by situating these contributions within the broader 6G standardization trajectory.