Towards New Approaches in Machine Consciousness Research

Abstract

This article examines contemporary approaches to the study of AI consciousness and their methodological limits. The analysis is structured around Jonathan Birch’s classification, focusing on theory-heavy, theory-neutral, and theory-light approaches, while also treating multidimensional and multilevel strategies as a distinct line of inquiry. The starting point of the article is a position of positive agnosticism regarding the possibility of machine consciousness. It is argued that, despite their significant differences, most existing approaches do not sufficiently problematize the transfer of concepts and methods developed for the study of consciousness in biological beings to computational systems. On this basis, it is argued that the central difficulty in the contemporary research lies in the insufficient justification of the aforementioned transfer. The lack of empirical data or the absence of a reliable test, while significant, do not constitute the key difficulties. Particular attention is paid to computational functionalism, which is either explicitly adopted as a presupposition, implicitly built into the framework of inquiry, or made the object of criticism. In these circumstances, multidimensional and multilevel approaches prove most productive, as they allow one to localize key methodological difficulties and reframe them as more specific research tasks, without claiming to offer a definitive resolution of the problems. The conclusion is that the limits of current research on machine consciousness are shaped to a large extent by an inherited conceptual framework; accordingly, the article proposes a general research orientation for further work on the problem.