Verified physical-world ground truth becomes AI's scarce input
As reasoning trends toward free, the binding constraint on useful AI shifts from model quality to the quality of its inputs. Verified, non-toxic, embodied ground truth — the kind that must be measured in the physical world, not scraped — becomes the scarce, defensible asset. This is RDKG's central wager and the reason every garden's verified data compounds in value.
The thesis follows directly from two verified machine-layer signals — inference cost collapse (~280x in 18 months) and data exhaustion (public human text runs out ~2026-2032): if reasoning gets cheap while the cheap training substrate runs dry, models converge and the differentiator moves upstream to the inputs.
Confidence 0.7, not higher, because the falsifier is live: rapid progress in synthetic data and simulation could narrow the gap for some physical tasks. The bet is hedged toward domains where the world is irreducibly messy — materials, health, growing things — which is also where the spine lives.
Treat every garden's verified dataset as a balance-sheet asset and build the Machine Lane (llms.txt + MCP + agent pricing) to monetize it.
domains: AI · data-moat · economics-of-intelligence · machine-layer