ABSTRACT
This chapter describes how to approach developing introspection-based artificial intelligence (AI). It examines several questions that remain outstanding about the recommended approach, and how to go about doing introspection more “correctly” than some past researchers have done. Introspection can be refined in the sense of adding details, sometimes even changing the fundamentals of the picture. There is every reason to expect that any known issues of reliability in testimony would arise in the case of introspection, so it is worth summarising one of the canonical sources on the technique of writing history, “The Historian’s Craft”. Pristine introspection is better because the pollutions come from cultural artefacts, and we are aiming specifically for anthropic AI. In AI, some of the more basic skills may be implemented using an introspection-based algorithm, some can be implemented using some other AI, and some can just be hard-coded.
