Who do you trust? Teaching in the age of AI

I’m in Mumbai for EMCON 2026, the national conference of the Society for Emergency Medicine India. This is their 28th conference, with colleagues from India and overseas coming together for three days of emergency medicine, from 30 October to 1 November. It’s a real treat to have been invited to join them.

We’ve been here for a few days, hosted by Nitin and Charita Jagasia, both doctors in Mumbai and former colleagues from Southampton. They have looked after us amazingly well. Nitin is an emergency physician and EMCON’s organising secretary, so he has rather a lot going on this week. He’s still made time to check we’re coping with the sensory overload that can come with a first visit to India.

My talk today is about how we help doctors become ready for independent practice. Preparing it has been a really welcome chance to stop and think, not just about how we decide that colleagues in training are ready to practise independently, but also about where technology may, or may not, fit into that process.

It has also made me think about trust in both directions. We need enough evidence to trust a trainee with increasing responsibility. They need to trust us enough to say when they are unsure and ask for help.

Where AI can help doctors learn

I would certainly classify myself as a cautiously optimistic early adopter of technology in medical education. We started the St Emlyn’s podcasts before it was trendy for everyone to have one. I truly believe that simulation can offer great learning, with the right facilitator, and now we have AI.

We can give someone a fictional patient, ask what they would do, then change a detail and see whether they change their mind. They can have another go without waiting for the next suitable patient to come through the doors. We need to check the material and the answers, but I can see plenty of reasons to be interested in that sort of practice.

There is even some encouraging evidence around AI. In a small trial published this month, people who practised serious-illness conversations with an AI patient and automated feedback improved more than those given reading material. The improvement was assessed in conversations with actors playing patients immediately after training. It was a short-term result in a particular task, and the study did not compare AI practice with interactive teaching from a person. Even so, it is a useful example of technology doing some actual teaching.

I can imagine using this to help a trainee rehearse a difficult conversation with an AI patient, then discussing with them what they tried and why. We could follow that with observation and feedback on a real clinical encounter. That is an approach I would like to explore; the trial did not test that combination.

Simulation can also help us assess what someone can do, as well as help them learn. There is evidence linking resuscitation assessments in simulation with those in the workplace, although the study was small and the relationship imperfect.

The problem would be assuming that providing all this means we have done our part. A doctor can have access to excellent resources but still need somebody to help them with a patient, discuss a decision they are unsure about, or tell them honestly how they are getting on. A recommendation to listen to a podcast doesn’t fulfil that need.

There still needs to be someone to call

I remember being a registrar, a while ago now, and feeling like a fool when a consultant recognised something I had missed, seemingly immediately. The consultant was on call from home, and I was struggling to treat a patient’s hypotension. I had been giving fluid with no improvement. I couldn’t see signs of sepsis, hypovolaemia, anaphylaxis or cardiogenic shock, but he just wasn’t getting better.

My consultant, who happened to be Cliff Reid, came in, took one look at the patient and used the new-fangled ultrasound machine. This was 2007. He immediately identified a pericardial effusion with tamponade physiology.

I hadn’t minded calling for help. I knew I needed it. What bothered me afterwards was that I thought I should have recognised what was happening, and I felt I had let him down.

I’m now the consultant receiving the call, but I think it’s worth remembering that feeling. We need to help with the patient, but the doctor who called us may also be worried about what we think of them. That doesn’t mean reassuring them that everything was fine if it wasn’t. It means taking the trouble to find out what happened, and explaining what we think they need to do differently.

Perhaps they knew the possible diagnoses but couldn’t decide which to pursue. Perhaps they understood the treatment but didn’t recognise when it wasn’t working. We need to take the time to hear their thought process before deciding how we can best help.

They should be able to ask about my thinking too. If I suggest a different investigation or change the plan, it is reasonable for the trainee to want to know why. Otherwise they have seen what I did without necessarily understanding how I arrived at it. And if they have a good reason for doing something differently, I ought to be prepared to listen.

All of this takes trust. A trainee needs to know that their senior won’t think less of them simply for asking questions or admitting that they don’t know the answer. But getting to know someone takes time. A trainee may be quite comfortable saying they don’t understand something to one consultant and rather less comfortable saying it to another.

Research on asking for clinical help has described how worries about appearing capable can affect those requests. And we certainly cannot assume that a doctor who asks fewer questions is necessarily becoming more independent.

I don’t think we can build that relationship simply by telling people we are approachable. How we respond when they interrupt us, particularly when we are busy, is part of it. So is coming back to a conversation we said we would have. We still have to be honest about concerns, but somebody should be able to ask for help without wondering whether the act of asking will be held against them.

We have to make time to do it

I know this is easier to write about than to do. I don’t know what your emergency department is like, but in mine there is always another patient to see. It can feel difficult to stop and talk when you can see how much work your colleagues still have to get through.

Sometimes a short explanation at the bedside is enough. Sometimes we need somewhere private and a bit longer, particularly if the trainee is upset or we are worried about their performance. Someone has to look after the other work while that happens. And we really cannot keep promising to find time later without making some arrangement for it. All too often I have done this and, to my regret, the conversation has never happened.

A recent survey of emergency medicine teachers and trainees identified time pressure and lack of privacy as barriers to feedback. Trainees also expressed a desire for more honest, specific feedback.

This is also part of being an emergency physician. A consensus framework aligned with the International Federation for Emergency Medicine’s curriculum identifies five core entrustable professional activities, including teaching emergency staff alongside patient resuscitation.

Those of us who run departments have to take some responsibility for this. Ten minutes spent properly with a registrar might mean I see one fewer patient myself. I don’t think we can keep asking consultants to teach while expecting their clinical workload to be unaffected. Giving the trainee more material to work through at home won’t solve the lack of time to work with them in the department.

Something to try on the next shift

This doesn’t mean we need to make every shift into a teaching programme, but we can look for opportunities to teach and plan ahead for them. I’d start with four steps:

1. Agree a learning need. Ask the trainee what they would like help with.
2. Watch them work. Choose one part of their work to observe that day.
3. Discuss it together. Ask about their thinking, explain yours and agree what to work on.
4. Arrange another opportunity. Make a plan to see how they get on next time.

As they become more capable, we can give them more responsibility. We will still have to decide what support they need, using what we and our colleagues have seen, and be available if things become difficult. A good performance in one setting is useful evidence; it doesn’t remove the need to think about the circumstances in which we are asking them to work.

I want us to keep making better use of technology. But when we decide how much responsibility a trainee can take on, we need to have watched them work and understood their thinking. They also need to trust that we will listen when they are unsure, help when they call and make time for the conversation afterwards.

References and further reading

Some of the reading behind the talk and this article, with a short note on what each paper adds.

1. Haut KG, Hasan M, Carroll T, Epstein RM, Sen T, Hoque E. Effects of a generative AI virtual patient on serious illness communication skills: randomized controlled trial. JMIR Medical Education. 2026;12:e93034.
   An example of AI-supported practice improving communication with simulated patients; it did not test readiness for independent emergency practice.
2. Weersink K, Hall AK, Rich J, Szulewski A, Dagnone JD. Simulation versus real-world performance: a direct comparison of emergency medicine resident resuscitation entrustment scoring. Advances in Simulation. 2019;4:9.
   Useful for understanding what simulation can contribute to assessment, and why it should be considered alongside workplace observations.
3. Kennedy TJT, Regehr G, Baker GR, Lingard L. Preserving professional credibility: grounded theory study of medical trainees’ requests for clinical support. BMJ. 2009;338:b128.
   Explores why asking for help can involve concerns about professional credibility as well as the patient’s needs.
4. Teng D, Venkataraman A, Singer A, et al. Entrustable Professional Activities for Emergency Medicine specialists. International Journal of Emergency Medicine. 2025;18:248.
   The consensus framework that includes teaching emergency staff among five core professional activities.
5. Chamberland M, Mamede S, St-Onge C, Setrakian J, Bergeron L, Schmidt H. Self-explanation in learning clinical reasoning: the added value of examples and prompts. Medical Education. 2015;49(2):193–202.
   Supports actively working through explanations of reasoning, based on diagnostic case tests with medical students.
6. Maneshi A, Fay A, Young M. Daily feedback in the emergency department: barriers and enablers. CJEM. Published online 19 March 2026.
   A survey of teachers and trainees about feedback, including limited time, lack of privacy and difficulty having honest conversations.
7. ten Cate O, Hart D, Ankel F, et al. Entrustment decision making in clinical training. Academic Medicine. 2016;91(2):191–198.
   A conceptual account of how supervision decisions depend on the trainee, supervisor, task, circumstances and their relationship.

More from St Emlyn’s

Making Feedback More Than Just Noise: Natalie May’s recorded talk about feedback in clinical practice.

Deliberate practice: the St Emlyn’s discussion of practice, feedback and improvement.