Abstract
This paper continues the series of theoretical reports on friendly AI
alignment. Friendly AI alignment is a failsafe alignment capable of
operating beyond Artificial General Intelligence threshold, where direct
control is not feasible and human safety is threatened. The alignment
class is of interest for Artificial Super Intelligence instantiation.
This paper presents an information system architecture that can host
friendly AI alignment. A sobriquet for the architecture is Serial
Comptroller Network (SCN). It implements a human-assisted,
uninterpretable approach. The discussion focuses on a minimalistic
solution within the framework of the proposed architecture. An
assessment for its potential, utility, deficiencies and further
improvement options is presented.