Making High-Level AI Design Decisions Explicit Using a Binary Stream System-Designation Approach

Abstract: Some crucial decisions in AI design tend to be overlooked or factor choices are assumed implicitly. The question often answered first is what the AI will do, not how it will interact with the rest of the world. This reduces our understanding of the possible types of AI that can be developed and their potential impacts on humanity. As an initial AI taxonomy, I present binary choices for 10 of the subjectively most separable and influential high-level design factors, then give brief examples of several of the 1024 possible systems defined by those choices. This supports a simple binary stream approach to system designation based on translating the stream of choices into decimal notation, giving a short-hand way of referring to systems with different properties that meet specialized needs. Further, underspecified or generic systems can be designated using the binary stream approach as well, a notational feature that supports modeling the impacts of AI systems with selected characteristics.