Constructing Meaning Across Substrates: Human Cognition, Artificial Intelligence, and the Problem of Cross-Substrate Understanding

Abstract

Abstract The rapid development of artificial intelligence has intensified a familiar question: can machines think, understand, or possess consciousness? These questions, however, often assume that the meaning of terms such as understanding, representation, and meaning is sufficiently stable across biological and artificial systems. This paper argues that such an assumption may be premature. Human cognition constructs meaning through a complex interaction among perception, prediction, memory, affect, embodiment, social interaction, and goal-directed behavior. Contemporary artificial intelligence systems, particularly large language models, construct highly structured representations through computational mechanisms that differ substantially from those of biological cognition. These systems can produce context-sensitive, coherent, and behaviorally sophisticated outputs without providing direct evidence that their internal processes correspond to human forms of experience or meaning construction. The resulting problem is not simply whether artificial systems can understand. It is whether human observers can reliably determine understanding when the underlying mechanisms by which meaning is constructed differ from those of human cognition. This paper develops the concept of a cross-substrate translation problem: the possibility that apparent semantic agreement between humans and artificial systems may reflect successful translation between different representational systems rather than the construction of equivalent meanings. Drawing on perspectives from cognitive science, neuroscience, artificial intelligence, information theory, linguistics, evolutionary theory, and philosophy of mind, the paper proposes a conceptual framework for analyzing meaning construction across heterogeneous systems. Rather than attempting to establish whether current or future AI possesses consciousness or genuine understanding, the framework identifies several dimensions relevant to cross-substrate comparison, including representation, grounding, integration, prediction, goal relation, self/world modeling, action, and feedback. The paper concludes that the central challenge posed by increasingly capable AI may not be determining whether machines reproduce human cognition, but developing evaluation frameworks capable of recognizing forms of intelligence and meaning construction that do not necessarily resemble their biological counterparts. Keywords: meaning construction; artificial intelligence; cognitive science; embodied cognition; large language models; understanding; consciousness; cross-substrate cognition; AGI; ASI Author’s Note This paper began with a relatively simple question: what does it mean for two different systems to understand the same thing? Rather than treating artificial understanding as a problem that can be resolved by comparing outputs, I attempted to examine the question from an interdisciplinary perspective. Neuroscience, cognitive science, artificial intelligence, information theory, linguistics, semiotics, philosophy of mind, and evolutionary theory each describe different aspects of cognition and meaning. I have tried to bring these perspectives together not to establish a single theory of understanding, but to examine where their explanations converge, where they diverge, and where our current criteria may no longer be sufficient. As a human researcher, I am also aware of an unavoidable limitation: I can only examine artificial cognition through human concepts, human language, and human methods of observation. This creates a fundamental asymmetry. We are attempting to understand potentially non-human forms of representation using cognitive categories that were developed to describe our own. At the same time, the technological trajectory may not wait for our conceptual frameworks to become complete. The systems we are likely to encounter in the future may be more capable, more persistent, more autonomous, and potentially more difficult to interpret than the systems we study today. This creates a gap between what technology may become and what humans are currently able to recognize or explain. I have therefore tried to approach that gap without assuming either that artificial systems will eventually reproduce human understanding or that biological cognition is necessarily the only possible form of meaningful cognition. Perhaps the more difficult question is what happens between these positions. If human and artificial systems construct meaning through fundamentally different processes, there may eventually be a point at which neither side can be adequately understood using the other's criteria alone. At that point, the objective may no longer be to determine which system possesses the "correct" form of understanding, but to develop a workable point of translation, accommodation, and mutual interpretability. I do not know whether such a point will be necessary, or whether it will even be possible. This paper is simply an attempt to ask the question before we reach that point.