The Technology Changes. The Duty of Care Doesn't.

Artificial intelligence may be new to clinical practice. The principle governing a doctor's responsibility for the technology they use is not.

Long before generative AI, clinical decision-support systems or machine-learning diagnostics entered healthcare, Irish law had already confronted a fundamental question:

"What happens when clinical practice relies on a system or technology, but that system does not tell the whole story?"

One of the most important answers came from Dunne v National Maternity Hospital [1989] IR 91.

A case clinicians already know — with renewed relevance

Dunne arose from the management of a twin pregnancy at the National Maternity Hospital.

At the time, the hospital's practice was to identify and monitor only one foetal heartbeat in a known twin pregnancy. The case concerned, among other issues, whether the failure to identify and monitor the second heartbeat amounted to negligence.

The Supreme Court's decision became foundational to the Irish test for medical negligence.

The significance of Dunne goes beyond obstetrics or foetal monitoring.

It established principles for assessing whether a medical practitioner has fallen below the required standard of care. Importantly, adherence to a general and approved practice will not necessarily protect a practitioner where that practice contains an inherent defect that ought to have been obvious to a practitioner exercising ordinary care.

That principle remains highly relevant when the "system" involved is no longer a foetal monitor, but an algorithm.

AI does not create a new standard of clinical responsibility

Imagine a clinical AI system reviews a patient's information and produces a recommendation.

The output appears plausible.

The system is approved for use within the organisation. It may have been procured centrally, integrated into existing software and routinely used by colleagues.

But something about the patient does not fit.

The examination findings are inconsistent with the recommendation. Relevant information was absent from the input. The clinician knows that the system has limitations relevant to this patient. Or the recommendation simply does not make clinical sense.

The existence of an AI recommendation does not remove the clinician's existing professional obligations.

The Medical Council's position on artificial intelligence in medicine makes this particularly clear: doctors remain responsible for their clinical decisions. AI should augment rather than replace clinical decision-making, and clinicians must appropriately scrutinise and interpret its recommendations while retaining their own critical thinking, reasoning and professional judgment.

"The system recommended it" is not a substitute for clinical judgment.

From the heart monitor to the algorithm

Dunne was not an AI case.

It should not be presented as though the Supreme Court decided anything about artificial intelligence.

Its relevance lies instead in the continuity of the underlying professional principle.

Medicine has always involved tools: monitors, laboratory tests, imaging systems, scoring tools, electronic records and clinical decision-support software.

AI introduces a more sophisticated tool — one capable of generating recommendations, classifications, summaries and predictions that can appear highly authoritative.

That makes meaningful clinical oversight essential.

One recognised risk is automation bias: giving excessive weight to an automated recommendation because it has been produced by a system perceived as sophisticated or objective.

The EU AI Act expressly recognises this issue. For high-risk AI systems falling within its human-oversight requirements, Article 14 addresses the need for people overseeing such systems to understand their capabilities and limitations, remain aware of possible over-reliance on outputs, correctly interpret those outputs and, where appropriate, disregard, override or reverse them.

The regulatory terminology is new. The clinical principle is familiar.

Clinical oversight means more than having a human in the loop

A clinician clicking "accept" on an AI-generated recommendation technically involves a human.

That does not necessarily amount to meaningful oversight.

Effective oversight requires the clinician to remain capable of asking:

"Does this output make sense for this patient?"

That can require considering:

• 01whether the AI had the correct and complete information;
• 02whether important clinical context was excluded;
• 03whether the system is appropriate for this particular use;
• 04whether known limitations are relevant to the patient;
• 05whether the output corresponds with the clinician's own assessment; and
• 06whether a discrepancy requires further investigation.

This is not fundamentally different from questioning an unexpected laboratory result, recognising artefact on a monitor or reconciling an imaging report with the patient's clinical presentation.

The interface has changed. The responsibility to think has not.

Competence now includes understanding the tool

Another familiar professional obligation becomes particularly important when AI enters clinical practice: competence.

Section 94 of the Medical Practitioners Act 2007 provides for the maintenance of professional competence by registered medical practitioners.

The Medical Council's guidance on AI connects professional responsibility with understanding these systems and their appropriate use.

This does not mean that doctors need to become machine-learning engineers.

A clinician does not need to know how to manufacture a CTG machine to use and interpret it appropriately.

Similarly, a clinician does not need to build an AI model to understand what it is intended to do, what information it relies upon, its important limitations and when its output should be questioned.

The same professional framework still applies

AI can make healthcare faster and, when appropriately designed and used, potentially better.

It can identify patterns, support clinical decision-making, reduce administrative workload and provide another source of information.

But introducing an algorithm into the consultation does not displace the existing relationship between doctor and patient.

The Medical Council's Guide to Professional Conduct and Ethics for Registered Medical Practitioners continues to place patient safety, professional competence, clinical judgment, communication, confidentiality and good medical practice at the centre of the doctor's role.

Its approach to AI builds upon those principles rather than replacing them.

That may be the simplest way to understand responsible clinical AI:

AI is a new tool operating inside an established professional framework.

Clinicians already know how to work within that framework.

Use appropriate tools. Understand their limitations. Consider the patient in front of you. Question results that do not make sense. Maintain appropriate records. Protect confidentiality. Maintain professional competence. And ultimately exercise your own clinical judgment.

The technology changes.

The duty of care doesn't.