Definition
Plain language
A cheap stand-in for an expensive evaluation, used to guide search or selection.
As stated in the literature
A simplified approximation to a costly objective used during search or optimization (e.g., a O(1) replacement for a full simulator in STAR); a common locus of fabricated implementation claims in AI-written systems papers.
Why it matters: It makes large-scale search tractable, but if the surrogate silently disagrees with the real objective, results can look great on paper while being wrong in practice.
For example, instead of running a full physics simulation, a search algorithm might use a fast neural-network estimate of the simulator's output to score candidates.
Heard on the show
“… described in the paper as using bitwise integer encoding for conflict detection, an O-of-one surrogate cost model, and equidistant placement of high-contention anchor transactions in some carefully reasoned …”Episode 089 — When AI-Written Papers Read Well But the Evidence Underneath Is Broken