Legitimate question.

How is the act of an AI company downloading a copyrighted work and adding it into their “dataset” to generate a summary different from an individual downloading a copyrighted work and adding it to their “dataset” and writing a summary?

Both instances require consuming the material in some way. Both instances generate something new, transformative of the original work.

Why do AI companies get to torrent the entirety of human knowledge, but if any single person does it… Well we know what happened with Napster. Limewire. Kazaa. Megaupload.

Because to me, that hints at a flaw in your logic. AI companies are violating copyright. They had no permission to consume the copyrighted works.