the ai wars are loud right now.
authors are suing. publishers are suing. record labels are suing. the NYT is suing. every creative class is convinced its work has been stolen, and every model lab is convinced training is fair use.
the arguments collapse into two camps. either training is theft and must stop, or training is reading and there is nothing to discuss.
both miss what is actually happening.
here's the analogy i can't unsee: we already have machines that can memorize books — humans.
some savants can read a novel once and keep it verbatim for decades.
nobody calls that infringement; law and ethics don't police what happens in your head.
the line is when you externalize it: when you write it back out and hand it around as a substitute copy.
training an ai is the reading step — absorption, compression, interiorizing.
the real harm is the dump: when a prompt turns the system into a photocopier for protected paragraphs.
so aim the fight where it belongs: test and punish verbatim reproduction, build guardrails against it. keep the right to learn, without quietly granting the right to reprint.