The Copernicus Problem: Time as the Invariant Denominator of Value in the Age of Artificial Intelligence

This paper argues that the SI second — defined by the fixed frequency of a cesium-133 atom — provides the natural invariant unit of account for pricing artificial intelligence. As AI inference becomes a commodity market, traditional monetary units fail to capture the rapid deflation in the cost of useful intelligence. We propose the Capability-Second as a composite measure: the amount of useful AI output (scored on standardized tasks) delivered per second of latency, per dollar of cost. This unit is grounded in physical metrology rather than monetary convention, making it robust to inflation, currency fluctuation, and provider-specific pricing distortions.

We introduce the Capability-Seconds Index (CSI), a daily benchmark that operationalizes this framework across frontier AI models. Using empirical measurements of 16 models evaluated on identical tasks, we demonstrate a 287× efficiency spread between the cheapest and most expensive models — a gap that reflects pricing dispersion in an immature commodity market, not capability differentiation.

The paper draws on four intellectual traditions — time-banking, compute-as-currency, the energy theory of value, and Bitcoin’s proof-of-work — to argue that legibility (commodity compute pricing) enables denominability (a standard unit), which enables invariance (the SI second as anchor). We present the CSI as both a measurement instrument and a test of this thesis: if AI inference is truly deflating along a measurable curve, the index will show it.