AUTHOR=Spasokukotskiy Konstantyn 

TITLE=Synthetic consciousness architecture

JOURNAL=Frontiers in Robotics and AI

VOLUME=Volume 11 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2024.1437496

DOI=10.3389/frobt.2024.1437496

ISSN=2296-9144

ABSTRACT=This paper presents a theoretical inquiry into the domain of secure artificial superintelligence (ASI). The paper introduces an architectural pattern tailored to fulfill friendly alignment criteria. Friendly alignment refers to a failsafe artificial intelligence alignment that lacks supervision while still having a benign effect on humans. The proposed solution is based on a biomimetic approach to emulate the functional aspects of biological consciousness. It establishes “morality” that secures alignment in large systems. The emulated function set is drawn from a cross section of evolutionary and psychiatric frameworks. Furthermore, the paper assesses the architectural potential, practical utility, and limitations of this approach. Notably, the architectural pattern supports straightforward implementation by activating existing foundation models. The models can be underpinned by simple algorithms. Simplicity does not hinder the production of high derivatives, which contribute to alignment strength. The architectural pattern enables the adjustment of alignment strength, enhancing the adaptability and usability of the solution in practical applications.