This study proposes a conceptual model that examines the physical cost of meaning production by addressing the distinction between information and meaning within a thermodynamic framework. Today, artificial intelligence systems are trained on trillions of parameters and process vast amounts of data; however, data abundance does not always translate into proportional value generation. The central claim of this study is that information (syntactic layer) alone does not generate value, whereas meaning (semantic layer) emerges in relation to context, agency, and utility, and that this process has a measurable energy cost. In this context, the Δ(E + I + A) model is proposed. The model represents the dynamic transformation over time of the components Energy/Effect (E), Information (I), and Meaning/Agency (A). While a conservation-like balance at the global scale (Δ(E + I + A) ≈ 0) is assumed, in local open systems the meaning gradient (∇A ≠ 0) is defined as the driving force of learning, adaptation, and creativity. Through an extension of Landauer’s principle to the semantic layer, the study introduces the concept of the “s-bit” (semantic bit) and argues that meaning gain (ΔA) is associated with energy consumption and the export of environmental entropy. The article presents testable experimental hypotheses such as the semantic overhead hypothesis and the entropy–meaning trade-off law; in particular, it proposes optimizing the energy per meaning (s-bit/joule) metric in artificial intelligence training, thereby introducing a new research agenda for sustainable and efficient information systems. The study does not claim to establish a physical law; rather, it offers a conceptual and operational framework for analyzing meaning production in human-centered open systems.
Serdar Çakıroğlu (Sat,) studied this question.