Indeed! And even subjectively, I find it difficult to believe that even if AI-written, the new information learned has negative marginal utility (or less marginal utility than had it been human-written). That is in fact the basis of the point I am making, though it is implied rather than stated.