How AI Can Assist Medical Device Engineering Without Replacing Engineering Judgment

What AI is good at

• Extracting candidate information from unstructured text
• Normalizing data from different sources into a consistent format
• Drafting document sections from an existing structure
• Checking consistency of naming, format, and relationships
• Flagging possible gaps or conflicts across large record sets

What should not be delegated

Risk acceptability, verification adequacy, regulatory pathway decisions, and formal approval require qualified people making judgments on specific evidence. AI can organize information and propose candidate relationships, but it cannot carry responsibility for those decisions.

A recommended workflow

• Input: public or authorized data with explicit source and scope
• AI assistance: extraction, normalization, relationship suggestions, drafting
• Reviewable output: preserve sources, confidence, and open questions
• Human review: confirm, correct, and add evidence
• Approval: completed by authorized personnel with a review record

Common failure modes

• Plausible content that cannot be traced to a source
• Presenting a candidate relationship as a confirmed conclusion
• Treating automation as a substitute for evidence quality
• Hiding uncertainty or failure cases
• Publishing regulated documents without human review

Practical principles

Place AI in the role of assisting organization and drafting, not independent decision-making. Every important output should retain its source, status, and review record. Every unverifiable conclusion should be marked rather than smoothly written into a document.