The General Constraining Arguments of Artificial General Intelligence and the Prospects of Robust Alignment

A discourse on the current state and possible scenarios of the development of artificial intelligent systems toward the level and abilities of general intelligence has been ongoing actively. One of the areas that has drawn attention from researchers is the possibility and challenges in the alignment of artificial intelligent systems including those advancing beyond the current state-of-the-art toward human-like cognitive abilities with human norms and values. While practical activities, policies and projects aimed at ensuring alignment with the current generation of intelligent systems have been well underway, in this study we focus on a more general question: what are the prospects, chances and/or the possibilities of ensuring robust alignment of secondary intelligent systems, in the sense defined in the study, in the further horizon and to the level of general intelligence. To advance in answering this question we attempted to determine and describe the minimal necessary characteristics that intelligent systems progressing toward that level must possess with the conclusion that under some reasonable assumptions, developing secondary intelligence capable of achieving the level of general intelligence will be able, at some stage of their cognitive progress, to form their own views, intents and objectives. The conclusion of our analysis is that the scenarios of “runaway” general intelligence that has advanced beyond the possibility of reliable control and perhaps, even comprehension, cannot be excluded for intelligent sys-tems advancing toward the level of general intelligence and has to be exam-ined closely and rigorously by the research community, policymakers and general society