- AI literacy
- Knowledge and skills that help people understand AI's uses, limits, opportunities and risks in their role.
- AI management system
- The organisation's policies, responsibilities and processes for governing AI.
- Assurance
- Evidence and assessment used to judge whether relevant requirements and safeguards have been met.
- Automation bias
- Giving an automated recommendation undue weight instead of examining the available evidence.
- Calibration
- How closely estimated probabilities match observed outcomes in the population being assessed.
- Clinical validation
- Evaluation of whether a system performs appropriately for its intended clinical purpose and population.
- Deployer
- A person or organisation using an AI system under its authority; the exact legal definition depends on the relevant framework.
- DPIA
- Data protection impact assessment: a structured assessment of risks that processing creates for people's information rights.
- Drift
- A change over time in data, context or relationships that can alter the performance of a system.
- External validation
- Testing on data from a setting, period or population outside the model’s development dataset. A random hold-out from the original dataset is internal validation, not external validation.
- Generative AI
- AI that produces content such as text, images, audio or code.
- Hallucination or confabulation
- An AI output that presents invented or unsupported information as if it were factual.
- Hazard log
- A maintained record of hazards, possible harms, assessed risks, controls, responsible people and outstanding issues.
- Intended purpose
- The use a product is designed and claimed for, as described through relevant instructions, labelling and promotional material.
- Interoperability
- The ability of systems to exchange and use information appropriately.
- Model update
- A change to the model or related product that may alter behaviour, performance or scope.
- PCCP
- Predetermined change control plan: a defined plan for specified device changes, their evaluation and implementation.
- Provider
- An entity supplying or developing an AI system in a defined role. This can mean something different from a healthcare provider.
- Safety case
- An evidence-supported explanation of why a particular system and deployment are considered acceptably safe.
- Sensitivity and specificity
- Sensitivity concerns identifying people who have the target condition; specificity concerns identifying those who do not.
- Special-category data
- Under UK/EU data-protection terminology, information given additional protection, including health data.
