Shooting from the hip into our own foot? A perspective on how artificial intelligence may disrupt medical training

Editor – I have enjoyed reading articles in the Future Healthcare Journal and wider medical literature about the potential for artificial intelligence (AI) to positively transform healthcare and improve patient outcomes. Recent articles have highlighted prospects of AI reducing administrative burdens, improving diagnostic accuracy and working synergistically with robotic technologies.^1,2 However, risks and uncertainties surrounding AI warrant a cautious approach to its implementation. One commonly overlooked risk is disruption to medical training, which fundamentally relies on experiential learning to refine decision making and improve situational judgement. Medical school and early education focuses on knowledge acquisition, which is essential but not by itself sufficient to prepare a doctor for clinical practice. Following graduation from medical school, clinicians rely on practising their decision making, gaining experience and learning from it.

As AI begins to exert its effects on the medical field, junior and senior clinicians will be affected differently. The job of a junior doctor typically consists of some automatable routine work eg evaluating patient records, simple diagnoses and paperwork. On the surface, this repetitive work may appear undesirable, but it is crucial to the experiential learning model and is a key component of junior doctor training. Given a long enough timeframe, this routine work will become more efficiently delegated to AI, which can work faster, more efficiently and for longer hours. This may result in a reduced demand for those junior doctors, whose work has been substantially altered.

On the other hand, there will always remain a need for specialist consultants to maintain control over AI systems, to refine them and to work synergistically with them. In fact, we may even see an increased demand for these specialists when the capacity of healthcare systems grows, as a result of operational efficiencies provided by AI. This becomes problematic if the career progression of junior doctors has been hindered. Since the jobs of senior specialists are relatively resistant to automation compared to trainees, we may see staff shortages for these positions in the long term.

There is also a threat of overdependence on AI if doctors are ill-equipped with the programming skills required to handle the technologies in clinical practice. Doctors must be able to understand, communicate and correct the outputs of AI systems. Without an understanding of dataset validation, algorithmic biases and machine learning principles, this seems difficult to achieve. General Medical Council guidance for UK medical schools currently makes no reference to computer or programming skills despite the fact that, in the future, these are likely to be of use to current trainees.³ Unless doctors are aware of the limitations of AI, there is a risk that they may become susceptible to automation bias, whereby they place more trust in the technology than their own judgement.⁴ Headlines are already emerging about AI outperforming doctors, who in the future may be reluctant to overrule these high-performing systems if they do not understand how they work.

While this letter has focused on illustrating a rhetorical view of disruptive implications of AI, it is worth noting that there are some AI systems, such as generative adversarial networks, which may substantially enrich medical education. On balance, careful consideration is required during the early stages of AI development because of the path-dependent nature of its progression. The timelines for these developments are uncertain, but this does not mean that they are impossible, or indeed far away.