[npunt]

Procedural generation is going to be fantastic for personalizing language and explaining ideas; it's pretty obviously the future for anything written that is learning & information oriented.

However I'm concerned these personalization systems will be too accommodating and just tell people what they want to hear. One way people use search is for motivated reasoning: I believe something, I search for it, I find confirming evidence. A procedurally generated system I imagine is especially prone to this kind of massaging of queries to get desired outputs - tweak a few input parameters in the form of a query, and out pops the answer you seek. It's a hard problem to test for.

Funny thing is when people search to find what they want to hear, they're often driven to content farms, and those are increasingly ML-generated. Seems this is just cutting out the middlemen!