I find your assumption dubious just based on recent computing history. The masses of today did not find digital voice assistants, search engines, or other technology which has obvious privacy concerns to be any kind of problem. In fact, it's been the opposite where the technologies have become commonly accepted and the average user doesn't seem to care in the slightest what's done with their information. The technologies proved to be more convenient than worrisome and "AI" seems to be no different. Kids are already using it to do their school work, so the idea that future generations will be apprehensive towards it seems unlikely.
The capability and power cost per use will also go down in time. If there are some really good and common use cases, dedicated hardware will be produced so that those models can operate quickly and at low cost in much the same way smartphones have dedicated video encoder/decoder built in to the SoC because playing and recording video is a common task that would be disproportionately expensive to do with their CPU cores. I don't know if anyone knows exactly what those particular uses will be right now, but there are plenty willing to speculate or who have an idea to try out.
You also have to consider the opportunity cost. Even if the AI is just used for something inconsequential such as making memes or other images to shit post on social media, the AI can spit out something good enough almost immediately. That does have some energy cost associated with it, but it may be less than what it would take a human to fire up Photoshop or some other image editor to make it manually. Like any other tool it can be used productively to save the time and labor of many humans or it could be used for no particular purpose beyond amusement or enjoyment.