Deterministic AI is an architecture for regulated decisions in which the determination is computed by executable rules over structured facts — producing the same verdict from the same facts, every time — while probabilistic models handle what they are good at: extracting structure from unstructured reality. The property purchased is repeatability under scrutiny.
The division of labour is precise. A language or vision model reads the drawing, the filing, the contract — and extracts entities, dimensions, and amounts with a confidence attached. A rules engine then evaluates machine-executable logic against those facts, citing the governing provision for every verdict.
Deterministic evaluation is increasingly written into procurement for compliance automation, because a permit, a filing, or an audit finding that varies between runs is not a defensible determination — however intelligent the model that produced it.
This vocabulary is implemented, not aspirational — it describes how the MAIA substrate actually runs. See the platform or read the research.