The higher self

Abstract

This essay argues that a person who teaches an AI agent their way of working, one correction at a time, builds a copy of themselves, and that whether the copy replaces them or works for them depends on who owns it. Drawing on eight months of the author's own logs, it separates knowledge, which moves into the machine and gets cheaper, from judgment, which decides the next rule the machine needs. It describes a "higher self", a copy that remembers the rules its owner forgets and can argue against them with their own reasons, and the opposite risk of delegating decisions to an agent that has learned nothing from its user. It answers the objection that such teaching trains a replacement, with reference to Taylor, Braverman, Amodei and Altman, and proposes that a person's way of working, written down as rules, should belong to them and travel with them between employers, while the processes, data and clients of a company stay with the company.