Human Oversight of Clinical AI: Automation Bias and Safe Override

Program overview

Meaningful human oversight is now a formal expectation of clinicians working alongside AI, and it is much harder than it sounds. When a system is right most of the time, people stop checking — this is automation bias, and it is well documented in aviation, radiology and clinical decision support. This course teaches the discipline of oversight: knowing what the tool can and cannot see, recognising the situations where it degrades, maintaining independent clinical reasoning, exercising override authority, and documenting the disagreement. Evidence suggests around a quarter of clinicians cannot reliably identify a clinically invalid AI output unaided.

Why this program

Who it is for

Program objectives

• Define meaningful human oversight of an automated clinical system
• Explain automation bias and the conditions that produce it
• Identify the situations in which an AI tool is likely to degrade
• Maintain independent clinical reasoning alongside an AI recommendation
• Exercise and document an override of an AI output
• Escalate a suspected systematic AI failure

Learning outcomes

• Identify automation bias in a described clinical interaction
• Evaluate an AI recommendation against independent clinical reasoning
• Recognise conditions under which a model's performance is likely to fall
• Apply an override decision and document it defensibly
• Differentiate a one-off error from a systematic model failure
• Plan an escalation route for a suspected AI safety issue

Curriculum

Duration and structure

6 lessons across 6 modules. Duration is expressed in study hours.

What you receive

Delivery format

• Self-paced — start any time.
• Study on any device.
• Lifetime access to the version you enrolled in.

Program handbook