Healthcare AI
Artificial intelligence systems applied to health processes, data and services. Value does not equal automation: it depends on purpose, data quality, oversight and accountability.
A—ZCRITICAL GLOSSARY
Words create the future.
First they must mean something.
Operational definitions for healthcare AI, longevity, data and governance. Written to clarify, not impress.
AI literacy
The knowledge and capabilities required to understand AI opportunities, limits and risks within one’s role. It is not one course for everyone, but role-appropriate competence.
Human oversight
The practical ability of a competent person to understand a system, detect anomalies, challenge its output and intervene before automation becomes uncontrolled decision-making.
AI governance
The rules, roles, controls and accountabilities surrounding a system throughout selection, deployment, use, monitoring and retirement.
Use case
A bounded description of the problem, users, data, output and supported decision. Without a precise use case, “we use AI” means very little.
Model drift
A gradual loss of fitness caused by changes in data, populations, processes or context compared with those in which the system was tested.
Bias
A systematic distortion capable of producing unfair or unrepresentative results. It may originate in data, labels, process design or the way outputs are used.
Explainability
The possibility of understanding relevant aspects of how a system works or why it produced an output. It does not always amount to a clinically useful explanation.
Longitudinal data
Observations collected over time about the same person or system. They support the reading of trajectories and change rather than isolated snapshots.
Longevity technology
Tools supporting prevention, awareness, autonomy, physical activity and continuity of information across the life course.
Intrinsic capacity
A WHO concept describing the combination of an individual’s physical and mental capacities. It is a lens on functioning, not a single magic score.
Wearable
A device worn on the body that collects signals or contextual information. Sensor accuracy, adherence and interpretation are separate problems.
Weak signal
A seemingly marginal piece of information that may precede meaningful change. Separating it from noise requires context, history and judgement.
Interoperability
The ability of different systems to exchange data while preserving meaning, provenance and usability. Connecting two APIs does not guarantee mutual understanding.
Human-in-the-loop
A design in which human contribution is an explicit part of the process. A final signature is insufficient: people need authority, time and competence to intervene.
Healthcare cybersecurity
Protection of data, devices and operational continuity. In healthcare, security protects not only information but the ability to work at all.
Digital twin
A continuously updated digital representation of an object, process or system. When applied to people, the term requires caution: a person is not reducible to a data stream.
Evidence gap
The distance between what a technology promises and what has been demonstrated in the setting where it will actually be used.
Adoption
The process by which people and organisations incorporate technology into real work. Installation and adoption are not synonyms.
Algorithmic accountability
The ability to assign decisions, controls, errors and corrections to identifiable roles. An algorithm cannot become the place where responsibility disappears.
A conversation, not a form