AI Economics

Concept

Non-rival ideas

An idea is non-rival when one person's use does not reduce the ability of others to use the same knowledge at the same time.

Explanation

A machine, worker, or unit of compute can normally serve only a limited number of uses at once. A design, formula, or algorithm is different: once known, it can guide production in many places simultaneously. This makes accumulated knowledge a powerful source of increasing returns and long-run growth.

Non-rival does not mean freely available. Patents, secrecy, copyright, access controls, and technical barriers can make an idea partly excludable. Exclusion can finance discovery, but it can also restrict diffusion; complete openness spreads knowledge while making it harder for a private inventor to capture the full social benefit.

AI sharpens this distinction. Model weights, algorithms, and discoveries are informational and potentially reusable, while the compute, energy, experiments, and skilled attention needed to create or run them remain rival inputs. Making existing knowledge cheap to distribute is also different from reliably extending the frontier of knowledge.

Examples

  • An algorithm

    The same published training technique can be implemented by many laboratories without the first laboratory losing its own ability to use it.

  • Model access

    A model's design can be non-rival while access remains restricted and every inference still consumes scarce chips and electricity.

Sources