Folk Reducibility and AI-Complete Problems

Abstract

The idea of an “AI-complete” problem has been around since at least the late 1970s, and refers to the more formal idea of the technique used to confirm the computational complexity of NP-complete problems. In the more formal context, the technique of reducibility was used to transform one problem into another that had already been proved to be NP-complete.

Our presentation takes a closer look at what we call “Folk Reducibility”, as an approximation to reducibility, in order to try and improve coherence regarding what constitutes tough AI problems. We argue that the traditional AI-complete problems like “the vision problem” and “the natural language problem” are too vague. We provide examples of more precisely specified problems, and argue that relationships amongst them provide a little more insight regarding where and how valuable problem relationships might emerge.