Safe and fair AI in health care

Making sure that statistical, machine learning, and AI models used in health and care are safe, fair, explainable, and monitored, including checking for bias across patient groups and meeting medical device and AI regulation.

Original to this reference.

Levels

1. Awareness: You can:

explain how a model can harm patients, for example through bias, drift, or over-reliance on its output
explain why models need validating on the population they will be used with
2. Working: You can:

measure model performance for different patient groups and report any differences
document a model's intended use, training data, and known limits, for example in a model card
contribute evidence about a model to a clinical safety case
3. Practitioner: You can:

plan the validation, bias assessment, and live monitoring of a model with clinical colleagues
decide when a model needs retraining, restricting, or withdrawing
advise teams on when a model may be regulated as a medical device and what evidence it needs
4. Expert: You can:

set the organisation's standards for assuring and monitoring AI in care
lead the assurance of high-risk models before and after deployment
represent the organisation with regulators and partners on safe and fair AI

Closest ESCO skills

principles of artificial intelligence (broad match)
data ethics (broad match)

Contains ESCO v1.2.1 data (https://esco.ec.europa.eu/). © European Union. Reuse authorised under Commission Decision 2011/833/EU.

Role levels that expect this skill

Role level	Role	Band	Expected level
Trainee data scientist	Data scientist	4	Awareness
Associate data scientist	Data scientist	5	Awareness
Data scientist	Data scientist	6	Working
Principal data scientist	Data scientist	7	Practitioner
Lead data scientist	Data scientist	8a	Practitioner
Head of data science	Data scientist	8b	Expert
Senior machine learning engineer	Machine learning engineer	7	Practitioner
Lead machine learning engineer	Machine learning engineer	8b	Expert