The Engineering of Artificial Consciousness

Abstract

Artificial consciousness is often discussed as if consciousness were the endpoint of increasing cognitive sophistication: more language, more reasoning, more planning, and more self-report. This article develops a different hypothesis. It argues that the most elementary form of consciousness is not discursive thought but affective valuation: the felt significance of a system's own state as better or worse for its continued existence. Drawing on brainstem theories of arousal, affective neuroscience, homeostatic accounts of selfhood, and the free-energy principle, the article proposes that artificial consciousness should not be sought first in fluent language models, but in artificial agents with intrinsic needs, viable bounds, and internal variables that matter to their own persistence. On this view, a system becomes a plausible candidate for artificial feeling only when it must monitor and regulate multiple self-relevant needs, predict the consequences of action for those needs, and allow internally valenced states to shape behaviour in ways not reducible to externally imposed reward maximization. The article further distinguishes simulation of affective language from affective subjectivity, outlines a multi-marker strategy for testing artificial feeling, and discusses why apparently irrational or non-instrumental behaviour may sometimes provide stronger evidence of felt motivation than perfect task optimization. Finally, it considers the ethical implications of engineering artificial agents that might become moral patients. The central claim is cautious but consequential: a machine that merely processes information is not thereby conscious; a machine that must care about its own continued existence may be closer to the functional roots of consciousness.