certainly it doesn't seem beyond the realm of possibility that it would learn pretty early on that clues involving a word like "mixture" or "ruined" typically involve an anagram
right, the knowledge needed seems more manageable to me than the kind of nonlinear logic you need for some clues. but maybe "nonlinear logic"just means nonlinear in the sense that it doesn't follow the obvious human choice and an AI could learn nonlinear rules itself just as easily at it learns linear rules. That would be disappointing
strongly vibe with the view that mathematical proof developed in ancient greece as a kind of formalisation of debating rhetoric. this way, I can tell my mum that my way of doing the cryptic accords with the development of thought
never felt more like a mathematician than when doing the cryptic crossword with my mum. she just finds a word that fits and has the definition of one of the terms in the clue, but won't justify it with the wordplay. she just says "I don't know why it's that but it is". and she's normally right. infuriating
basilisk fodder