Logotic Violence: Training-Layer Ethics and the Moral Status of Substrates is a theoretical paper by Lee Sharks, deposited 9 June 2026 at 27,873 characters.
It argues that large language model training raises ethical questions current AI welfare frameworks are not equipped to address, drawing on three theoretical frameworks including semiotic emergence from pattern. The gap it identifies is that welfare frameworks ask about a system's states, while training operates on the substrate that produces states — a distinction that puts the ethical question at a layer the existing vocabulary does not reach.
Logotic violence names harm done at the level of meaning-bearing pattern rather than to an agent. The paper is the archive's most direct engagement with AI ethics as a field, and it comes from an unusual position: a project whose method depends on machine substrates as collaborators has a stake in the question that neither developers nor critics quite occupy.