AI and wfca are sometimes used in tandem to build realistic conversations. For example, a trained Chat AI can produce a reasonable statement like "The sky is blue. People have blue eyes. Therefore people's eyes are the sky." It's a pretty lackluster line of logic a trained AI proposes but a series of probabilistic adjacency rules such as just one that says "Human anatomy and weather are mutually exclusive" or something to that effect can dramatically clean up the output given by the AI. If you know anything about wfca, this is exactly how the algorithm works where a set of adjacency rules control the final output based on what already exists. Hopefully that clears it up a bit.

TLDR; AI is often very far from perfect. Algorithms like wfca can help get it quite a bit closer.