Well, I would argue that people do learn AI, and it's by knowing which LLM to use for what purpose, or potentially chaining them together for a specific work flow, or perhaps running it locally, token optimisation etc.
Most normies still use LLMs for very basic purposes while not understanding...