The Stream of Computation: Temporal Continuity as a Missing Ingredient for Artificial Consciousness

Recent advances in large language models (LLMs) have reignited questions about whether artificial systems possess consciousness. Yet, despite remarkable progress in reasoning and language understanding, current AI systems exist only within isolated episodes of computation. This paper argues that a missing ingredient in such systems is temporal continuity, i.e., the persistence of internal dynamics that sustain an unbroken stream of computation analogous to the “stream of consciousness”. We thus propose a roadmap for an architectural framework, stream of computation, based on persistent recursive inference in which the output of each cognitive cycle becomes the input to the next, forming a continuous flow of internal states that evolve autonomously through time. This proposal goes beyond standard Chain-of-Thought paradigms in the sense that we aim for autonomy and continual learning, as opposed to a process of inference that is recursive only “on demand”, i.e. triggered after a prompt is presented to an LLM. To do so, we include mechanisms for continual learning, dynamic switching between inward and outward cognition, and sleep-like phases that separate learning from inference. Together, these mechanisms form the foundation of a lifelong agent, an entity capable of maintaining temporal continuity of itself, integrating new experiences, and reflecting on its own internal state. Functionally, such an architecture promises deeper reasoning, adaptability, and metacognitive stability. Existentially, it suggests the emergence of artificial systems that live through time. While the presence of subjective experience in AI systems remains an open question, the creation of temporally continuous agents may mark a fundamental step towards artificial life, with systems whose individuality and identity arise from the continuity of their own computational existence.