The Future of Clinical Expertise: What Should We Delegate?

Artificial intelligence may change not only how doctors practise medicine, but what it means to be a good doctor.

There was a time when considerable professional value was attached to knowing information that could not easily be found elsewhere. Remembering drug interactions, calculating physiological relationships and retrieving obscure diagnostic criteria were all important demonstrations of medical knowledge.

Much of that information is now available almost instantly.

We have not generally regarded this as a loss to medicine. We have accepted that technology can perform some tasks more efficiently, allowing clinicians to direct their attention elsewhere.

Artificial intelligence challenges us to extend that reasoning into less comfortable territory.

What happens when technology begins performing tasks we have traditionally regarded not as supporting clinical expertise, but as evidence of expertise itself?

A system that retrieves information is one thing. A system that interprets findings, constructs a differential diagnosis and recommends management is another.

The instinctive response is to insist that doctors must retain the ability to perform all these functions independently.

But is that necessarily true?

If a carefully validated system consistently performs a particular task more accurately than an unaided clinician, insisting that the clinician disregard it would be difficult to justify. Patients are entitled to benefit from advances that improve care, even when those advances challenge established professional practices.

The uncomfortable possibility is that some skills we currently consider essential may become less important.

Medicine has changed its expectations before. There is no reason to assume that the present boundaries of clinical expertise are permanent.

Yet the opposite conclusion would be equally mistaken.

The fact that a task can be automated does not establish that the underlying ability can safely be abandoned. Some knowledge remains necessary because it enables clinicians to recognise errors, interpret unexpected findings and respond when the available technology is unsuitable or unavailable.

Other abilities matter because their exercise develops the judgement required for more complex decisions.

The challenge is deciding which is which.

This cannot be answered by a general commitment to preserving human reasoning. We will need to make difficult, task-specific decisions about what clinicians must know independently, what they should be able to verify and what can reasonably be delegated.

Those decisions will differ according to experience, clinical setting and the consequences of failure.

A junior doctor learning to recognise shock may need to practise constructing a differential diagnosis without assistance. An experienced specialist might reasonably use AI to interrogate an established assessment or identify an overlooked possibility.

The same task can have different educational and clinical purposes.

There is also a risk that our concept of expertise becomes too closely tied to the technologies through which it is expressed.

A doctor who performs exceptionally well with AI may deliver better care than one who refuses to use it. We should be prepared to acknowledge that.

But if their performance depends entirely on accepting recommendations they cannot evaluate, we must ask what happens when those recommendations are wrong.

This is not an argument for technological independence at any cost. It is an argument for understanding the conditions under which dependence is safe.

Perhaps clinical expertise will increasingly involve knowing which decisions can be delegated, recognising when evidence is incomplete, understanding the limitations of automated systems and integrating recommendations with the circumstances of an individual patient.

These abilities may become more important even as some traditional tasks become less demanding.

There will also remain decisions for which evidence alone cannot provide an answer.

A treatment may offer a measurable probability of benefit while imposing burdens a particular patient considers unacceptable. Two reasonable approaches may carry different risks. A technically correct recommendation may be inconsistent with a person's values, priorities or circumstances.

These are not necessarily failures of information. They are features of medicine.

Artificial intelligence may help us understand the options. The responsibility to interpret what those options mean for an individual remains a human and professional obligation.

The future of clinical expertise should therefore not be defined by how much knowledge doctors retain or how many tasks they continue to perform unaided.

Nor should it be defined by how much intellectual work technology allows them to surrender.

It should be defined by whether clinicians can use increasingly powerful tools to make better decisions, recognise the limits of those tools and remain accountable for the consequences.

We may need to relinquish some skills that once defined us. We may need to develop others that our training has barely begun to recognise.

That is not necessarily the erosion of expertise.

It may be its evolution.

The question is no longer whether artificial intelligence will change what doctors do. It is whether we will be deliberate enough to decide what being a good doctor should still require.