I like to think of AI as a university where I can draw up my own curriculum. A question about an artwork might send me into programming, and an experiment with code might lead me towards choreography. I can follow those connections while an idea is still taking shape, learning enough to attempt something and letting that first attempt reveal what I need to understand next.
That possibility is a large part of why I would encourage artists to work with AI. The distance between imagining something and knowing how to make it can be considerable. Finding the right course or collaborator takes time, and sometimes an idea is set aside before it has even been tested. AI can offer a way into a subject that previously felt beyond reach, giving that idea a chance to develop.
I find the difference between "show me how to play tennis" and "teach me how to play tennis" useful here. A demonstration can be a starting point; learning to play also involves practice and some understanding of why the ball keeps hitting the net. I still have to step onto a court. Similarly, when I ask a model for a piece of code, I want to understand it well enough to change something deliberately and see what happens. That gives me knowledge I can use in the next experiment.
I also want the time saved to remain available for exploration. An early test might tell me that an idea deserves more attention, or that I should leave it alone. Either outcome helps me decide what to do next. If starting again takes less time, I can afford to follow a possibility whose outcome is much less certain. There is room for a few more attempts to go nowhere.
That freedom depends on how carefully I work. My need for guidance is greatest precisely where my ability to judge the answer is weakest. An explanation can sound authoritative while giving me a poor account of its subject. I need to check sources beyond the model and test what it suggests, including with people who know more than I do. I also want to ask whose work has made this tool possible, and under what conditions it was used.
Abundance presents another difficulty. With a stream of polished options in front of me, I can drift into choosing whatever looks most accomplished. The ease of arriving at an image can make it tempting to settle there. I want to be able to explain why a particular result deserves further attention, and to set the whole batch aside if none does. That judgment takes work of its own.
I want AI to help me become more demanding of my own ideas. If it helps me get started sooner, I may choose to spend considerably longer finding out what is worth making.