Checking in after a couple of days I find several great new comments in the thread. Nice surprise! I'll try to address your points briefly.
One point raised by both Thistle Lion and Beige Bandicoot is the seeming enormity of the transformation needed to achieve what I propose (democratizing scientific understanding). I agree that it's a rather big thing to wish for, and something that cannot take less than at least a generation. However, I feel that in the essay I may have not done a great job explaining how I think this transformation could come about.
I mentioned in the essay that realigning incentives of educators and communicators would be helpful, and that scientists would need to refocus how they spend their time. But I don't believe these changes would be the first step, nor that they can happen in a (mostly) top-down manner. Instead, the ambition of the essay is to bring attention to this seemingly underappreciated opportunity, and in doing so spur more people, scientists or not, to take up the challenge. Just like graduate students today become enthralled with cutting-edge research problems, I hope that at least a fraction of them can be enticed by cutting-edge exposition problems. And similarly for the pop science community. I see it more as a grassroots transformation, and I think (hope) one that can fuel itself once it's started. I am trying to contribute in this myself, but saying anything more than that might give away my identity too early đŸ™‚
As for Persimmon Swan's argument about AI helping this cause, I am very attracted to this prospect, but I prefer to take it with a mild skepticism. I'm sure AI will help greatly in all the areas mentioned in your essay, so I believe that we agree on the general point. Still, my own experience with GPT 3/4 (and thanks for generating the summary!) is that they are very good at synthesizing, but not at all at understanding—not to mention making others understand. The "open exposition problems" require something much deeper than language skills. I've been surprised by AI advances many times in the past year, so it may or may not just be a matter of time.
