I’m going to assume that when you say “AI” you’re referring to LLMs like chatGPT. Otherwise I can easily point to tons of benefits that AI models provide to a wide variety of industries (and that are already in use today).
Even then, if we restrict your statement to LLMs, who are you to say that I can’t use an LLM as a dungeon master for a quick round of DnD? That has about as much purpose as gaming does, therefore it’s providing a real benefit for people in that aspect.
Beyond gaming, LLMs can also be used for brainstorming ideas, summarizing documents, and even for help with generating code in every programming language. There are very real benefits here and they are already being used in this way.
And as far as resources are concerned, there are newer models being released all the time that are better and more efficient than the last. Most recently we had Llama 3 released (just last month), so I’m not sure how you’re jumping to conclusions that we’ve hit some sort of limit in terms of efficiency with resources required to run these models (and that’s also ignoring the advances being made at a hardware level).