AI in Clinical Practice: What PLAB 2 Candidates Must Know in 2025

Artificial intelligence is no longer a futuristic concept confined to journals and conferences. It is actively reshaping how medicine is practised across the NHS, and every doctor — including those preparing for PLAB 2 — needs to understand the fundamentals. Whether you are explaining an AI-assisted diagnostic tool to a patient, navigating the ethics of algorithmic decision-making, or simply knowing where AI sits in your day-to-day clinical workflow, this knowledge is now part of the exam and the career that follows.

This guide brings together the high-yield concepts on AI in medicine that PLAB 2 candidates are most likely to encounter in OSCE stations, ethics discussions, and real clinical practice.

Why AI in Medicine Matters for PLAB 2

PLAB 2 is designed to test whether you can practise safely and effectively as a Foundation Year 2 doctor in the UK. The General Medical Council (GMC) and the UK Foundation Programme have made it clear that understanding digital health and AI is now a core professional competency, not a "nice to know" extra. AI tools are used in:

• Radiology reporting (e.g., stroke detection on CT, breast cancer screening)

• Pathology image analysis

• Sepsis and deterioration alerts (e.g., NEWS2-linked algorithms)

• Triage and streaming in emergency departments

• Decision support for prescribing and drug interactions

• Ambient scribe technology for clinical documentation

In your OSCE stations, AI may surface in three predictable ways:

1. A patient asks you about a tool or app that uses AI for diagnosis.

2. A scenario tests your understanding of the ethical principles governing AI use.

3. A station requires you to recognise the limits of AI recommendations and exercise clinical judgement.

The Core Concepts You Must Understand

1. What "AI" Actually Means in Clinical Settings

Most clinical AI in use today is narrow AI — algorithms trained to perform one specific task very well. Almost none of what is deployed in the NHS is general artificial intelligence. The most common subtypes you will meet in OSCEs are:

• Machine learning (ML): systems that learn patterns from large datasets (e.g., predicting AKI from blood tests and observations).

• Deep learning: a subset of ML using neural networks, dominant in image recognition (radiology, dermatology, ophthalmology).

• Natural language processing (NLP): powers ambient scribes, summarisation tools, and increasingly, patient-facing chatbots.

• Large language models (LLMs): the technology behind generative AI chatbots — relevant when patients ask about ChatGPT-style symptom checkers.

For exam purposes, you do not need to recite algorithms. You need to be able to explain in plain English what an AI tool is doing, what it is not doing, and why your clinical judgement still matters.

2. Strengths and Limitations

AI excels at pattern recognition in large datasets, consistency, and tasks that are repetitive or time-pressured. It is poor at contextual reasoning, rare presentations, and applying human values.

A useful exam framing:

A high-scoring OSCE answer would say something like: "This tool is a decision-support aid, not a decision-maker. I would still take a full history, examine the patient, and apply my own clinical reasoning before acting."

The Ethical Framework: GMC and NICE Guidance

The GMC has published clear guidance on AI and digital tools, and this maps almost perfectly onto the four pillars of medical ethics. Memorise the key points.

Beneficence and Non-Maleficence

• AI can improve care by reducing diagnostic delays, but it can also cause harm through automation bias — where clinicians accept AI outputs uncritically.

• You remain responsible for the decisions you make. AI does not transfer accountability.

• Validate that any AI tool used in your clinical area has appropriate regulatory approval (UKCA / CE marked, with MHRA registration where relevant).

Autonomy

• Patients have the right to know when AI is involved in their care.

• Consent discussions should include mention of AI tools where they materially influence diagnosis or treatment.

• Patients retain the right to refuse AI-influenced pathways if a reasonable alternative exists.

Justice

• AI can entrench health inequalities if training data under-represents certain populations (e.g., skin cancer detection algorithms performing less well on darker skin tones).

• Be alert to algorithmic bias and flag concerns through local governance routes.

• Equity of access to AI-driven care is an emerging NHS priority.

Professionalism (GMC Good Medical Practice 2024)

• You must understand the tools you use, including their limitations.

• You must not misrepresent AI outputs to patients or colleagues.

• You have a duty to report unsafe or unregulated AI tools.

AI Scenarios You Should Practise in OSCEs

Scenario 1: Patient Asks About a Symptom Checker App

Stem: A 28-year-old patient tells you they typed their symptoms into a chatbot and it suggested they have a brain tumour. They are anxious and asking what you think.

What the examiner is testing: Your ability to validate concern, correct misinformation without dismissing the patient, and explain AI limits.

Model answer structure:

1. Acknowledge anxiety and take the concern seriously.

2. Take a focused history and examine.

3. Explain that AI tools are not a substitute for clinical assessment — they can suggest possibilities but cannot weigh probability or context.

4. Reassure only after you have ruled out red flags; do not blanket-reassure.

5. Offer a safety net and clear return advice.

Scenario 2: A Colleague Relies Entirely on an AI Sepsis Alert

Stem: A nurse shows you that the AI sepsis tool has not triggered for a patient you are worried about. They suggest no action is needed.

What the examiner is testing: Your ability to exercise clinical judgement, challenge respectfully, and escalate appropriately.

Model answer structure:

1. Document your own clinical assessment (NEWS2 score, full set of observations, focused examination).

2. Explain that AI tools are supportive, not definitive — they have false negatives.

3. Escalate to your senior or the on-call medical team using SBAR.

4. Continue to monitor the patient closely.

5. Reflect afterwards on the importance of not deferring to an algorithm when your clinical instinct flags concern.

Scenario 3: Explaining an AI-Assisted Imaging Result

Stem: A patient has had a chest X-ray reported with AI assistance. They ask what "AI-reported" means.

Model answer structure:

1. Plain-English explanation: "An AI tool reviewed the image and highlighted areas the radiologist should look at carefully. A qualified radiologist still reviews and signs off the report."

2. Reassure that the final report is the radiologist's, supported by the tool.

3. Offer to explain the report findings once available.

4. Document the discussion.

Scenario 4: Capacity, Confidentiality, and AI Triage

Stem: A patient lacks capacity and a third party (e.g., a family member running a symptom app on the patient's behalf) requests clinical information.

Model answer structure:

1. Apply the Mental Capacity Act principles.

2. Confirm the patient's best interests involve a proper multidisciplinary discussion.

3. Do not share information with the family member without appropriate legal basis.

4. Explain that AI-generated information from a relative does not replace professional assessment.

5. Document the decision and your reasoning.

Talking About AI Safely in Stations

The best PLAB 2 candidates integrate AI knowledge naturally, without lecturing. Use these phrases fluently:

• "I would treat that AI output as a useful prompt, not a diagnosis."

• "My responsibility to the patient does not change because a tool is involved."

• "I'd want to understand what data the tool was trained on and whether it has been validated for our population."

• "If I had concerns about a tool's safety or bias, I would raise it through the clinical governance route."

Avoid jargon-heavy answers. The examiner is testing professionalism and judgement, not your knowledge of transformer architecture.

Common Pitfalls to Avoid

1. Treating AI as infallible — the moment you defer completely to an algorithm, you have lost marks.

2. Ignoring the patient in favour of the tool — AI discussion should always come back to the patient's concerns and consent.

3. Overpromising — do not claim AI can replace clinical assessment; the examiner will mark you down for this.

4. Skipping governance — if the station is about introducing a new AI tool, mention clinical governance, MDT approval, and audit.

5. Forgetting documentation — always document when AI has been used in decision-making, especially in serious diagnoses.

High-Yield References to Read

For PLAB 2 preparation, focus on the following accessible sources rather than deep technical papers:

• GMC: AI in healthcare guidance (general professional duties apply)

• NHS England: AI Deployment Guide and the NHS AI Lab resources

• NICE: Evidence standards framework for digital health technologies

• MHRA: Software and AI as a Medical Device guidance

• Royal College of Radiologists: position statement on AI in radiology

• Topol Review (2019) — still the canonical UK review on preparing the NHS workforce for AI

• WHO: Ethics and Governance of Artificial Intelligence for Health

Skim each for the key principles. You will not be examined on technical details, but you will be expected to speak intelligently about safety, bias, and accountability.

A Short Reflection Practice

Before your exam, practise a 60-second answer to the following question out loud:

"A patient asks whether you trust AI to make medical decisions. How do you respond?"

A strong answer acknowledges the value of AI in pattern recognition and efficiency, names automation bias and accountability as key concerns, and grounds the answer in the patient's right to a clinician who is thinking critically about their care.

If you can deliver that in under a minute, calmly and clearly, you are well prepared for AI-themed stations.

Final Word

AI in medicine is not a passing trend. It is a clinical reality that you will use, audit, explain to patients, and challenge when needed. The PLAB 2 exam is testing whether you are ready for that responsibility. By understanding the core concepts, the ethical framework, and the common station formats, you demonstrate the kind of safe, thoughtful practice the GMC expects from day one of your NHS career.

Practise these scenarios with a study partner, ideally with feedback on your tone, structure, and patient-centred language. The candidates who do well in AI stations are not the ones with the most technical knowledge — they are the ones who combine clinical confidence with clear, compassionate communication.