Research · Glossary
Glossary.
The terms our publications use, defined in plain language. Needless complexity is a form of opacity.
- Decision intelligence
- The discipline of supporting a specific decision, made by a specific person, with evidence that person can inspect, challenge, and stand behind.
- Verifiable evidence
- Evidence with a cited source that can be opened and checked. An unsourced specific is an unverified specific.
- Stated confidence (calibration)
- A confidence number is calibrated when it matches reality in the long run: of the claims marked 0.86, about 86 in 100 turn out true.
- Hallucination
- Content invented by a language model and delivered with confidence: names, figures, and citations are what models most confidently invent.
- AI governance
- The rules, roles, and safeguards under which an organization adopts AI systems: what they may operate, who answers, and how they are audited.
- Data governance
- The operating model that defines data quality, access, lifecycle, and ethics. The foundation of any AI or evidence strategy.
- Public algorithm register
- A published inventory of the automated systems an institution uses, with their purpose and safeguards.
- Effective human oversight
- Verifiable protocols that keep the decision with a person: who signs, what they review, and how disagreement is recorded. The opposite of automatic approval.
- Algorithmic transparency
- The ability to see how a system reached its result: reasoning, evidence, assumptions, and uncertainty, open to inspection.
- Noise (in decisions)
- Unwanted variability between decisions that should be alike: two officials, the same case file, different conclusions.