AI in medical devices · Medium · 13 min lesson
Human oversight and transparency: instructions for use, user interfaces and training
Regulation (EU) 2024/1689 Articles 13, 14 and 26, MDR Annex I section 23, ISO 13485:2016 clauses 6.2, 7.3.3 and 7.3.7, IEC 62366-1, GMLP principles 7 and 9
What you will learn
- Explain the five abilities Article 14 requires for effective human oversight and turn them into design inputs.
- Write instructions for use that meet Article 13 paragraph 3, including metrics, limits, misuse and oversight measures.
- Plan interface design, training and usability evaluation so that oversight is proven, recognise a weak record and answer the questions an auditor asks.
Key takeaways
- Human oversight is designed in: each Article 14 paragraph 4 ability needs a design input, owner and verification.
- Instructions for use carry metrics with intervals, limits, foreseeable misuse, oversight measures and log information (Article 13 paragraph 3).
- Interfaces should make doubt easy; avoid wording that sounds final and test wording with real users (IEC 62366-1).
- Deployers assign overseers with competence, training, authority and support (Article 26 paragraph 2); the provider states the assumptions.
- Prove it with a usability evaluation including wrong outputs, pre-set criteria, training records and override and suspension logs.
Free practice question
Corvane's first Pneumo-Flag screen shows a large green banner over the radiograph with the word normal. Pilot readers accept it without opening the plain image. The design team says the interface is final because it passed software verification.
What is the main weakness against Article 14?
Software verification does not show that the interface supports effective oversight; the banner invites over-reliance and needs a usability evaluation.
Correct. Article 14 paragraphs 1 and 4 point b require interfaces that support oversight and awareness of automation bias.
There is no weakness, because verification covers everything.
Not quite. Verification under clause 7.3.6 checks requirements, not whether users can oversee the system effectively as Article 14 expects.
The banner colour must be red instead.
Not quite. Colour alone does not address automation bias; Article 14 paragraph 4 point b concerns over-reliance in general.
Only the deployer is responsible for the interface.
Not quite. Article 14 paragraph 1 places the design duty on the provider, while Article 26 adds duties for the deployer.
Key terms
- human oversight: Measures that let natural persons monitor, understand, interpret, override and halt a high-risk AI system while it is in use (Article 14).
- automation bias: The tendency to rely too much on an automated output, especially a recommendation used for a human decision (Article 14 paragraph 4 point b).
- instructions for use: Concise, complete, correct and clear information for the deployer, with the minimum content in Article 13 paragraph 3.
- deployer: The organisation using the AI system under its own authority, for example a hospital (Article 3 of the AI Act).
- summative usability evaluation: A final test with representative users that shows the interface can be used safely, as described in IEC 62366-1.
- use error: A user action or omission that leads to a different result than the manufacturer intended or the user expected (IEC 62366-1).
Official sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Official Journal
- Article 13: Transparency and provision of information to deployers, AI Act explorer
- Article 14: Human oversight, AI Act explorer
- Article 26: Obligations of deployers of high-risk AI systems, AI Act explorer
- Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles (FDA, Health Canada, MHRA, 13 June 2024)
Go further
The full lesson has narrated slides, more scenario questions with explained answers and a simulation.
Open the AcademyWatch on YouTubeIllustrative example. Educational only, not legal advice.