ConsScale: a plausible test for machine consciousness?

Abstract

Is consciousness a binary on/off property? Or is it on the contrary a complex phenomenon that can be present in different states, qualities, and degrees? We support the latter and propose a linear incremental scale for consciousness applicable to artificial agents. ConsScale is a novel agent taxonomy intended to classify agents according to their level of consciousness. Even though testing for consciousness remains an open question in the domain of biological organisms, a review of current biological approaches is discussed as well as their possible adapted application into the realm of artificial agents. Regarding to the always controversial problem of phenomenology, in this work we have adopted a purely functional approach, in which we have defined a set of architectural and behavioral criteria for each level of consciousness. Thanks to this functional definition of the levels, we aim to specify a set of tests that can be used to unambiguously determine the higher level of consciousness present in the artificial agent under study. Additionally, since a number of objections can be presumably posed against our proposal, we have considered the most obvious critiques and tried to offer reasonable rebuttals to them. Having neglected the phenomenological dimension of consciousness, our proposal might be considered reductionist and incomplete. However, we believe our account provides a valuable tool for assessing the level of consciousness of an agent at least from a cognitive point of view.