Decisions and actions of all kinds by humans and AIs ultimately reduce to knowledge in use. The safety and effectiveness of actions, in particular, therefore depends upon the quality of knowledge used to inform and guide them. And while the use of knowledge for instrumental purposes occurs at the level of business processing, the production of knowledge as a precursor to action takes place in a different process called knowledge processing, in which distinctions between truth & falsity and legitimacy & illegitimacy can be made. Here, then, lies the difference between rules and rule-making rules.
Not all knowledge processes, however, are alike, much less effective in terms of the truth, legitimacy, or reliability of the knowledge they produce. Indeed, two AI systems endowed with different epistemologies can very well produce different answers to the same questions, some of which may even be contradictory. Choosing the best framework for use at the level of knowledge processing therefore matters greatly.
A third level of interest is the knowledge management (KM) level, the level at which management strategies and interventions aimed at improving knowledge processing, or rule-making rules such as the Susty Code itself, are devised. This is the 'applied epistemology' or meta-epistemic level. KM, that is, is knowledge process management — a management system for determining how best to kill our worst ideas before they kill us, and for addressing the alignment problem in AI. The higher the quality of knowledge in AI, the more aligned its outputs and actions can be with human values and well-being.
The Susty Code, then, is a game-changing innovation in high-performance fact/value knowledge processing — a creature of Epistemological AI — conceived at the level of KM with the quality of knowledge and business processing in mind. The epistemology it relies on is a variant of Critical Rationalism and takes the form of an integrated (fact/value), rule-making protocol for knowledge production in AI. No other approach to addressing the alignment problem in AI does this, and yet it must be done if AI is to be safe, ethical, and effective in the long run.
