Places where it is fine to fail, and places where we take on responsibility: Medical education in the age of learning with AI

Contents
  1. Knowledge, skills, and connections
  2. The ironies of automation
  3. Environments with rules and without
  4. Learning to take responsibility
  5. The doorman’s job
  6. References

With an AI simulated patient, you can practice breaking bad news alone, late at night, as many times as you like. Whether the patient cries or gets angry, no one is hurt. Being able to practice without tension and to redo the same scene again and again is a great advantage for learners.

At the same time, I suspect that the more we practice in places where failure has no consequences, the more frightening it may become to take on decisions in real situations.

Knowledge, skills, and connections

I think education has three elements: knowledge, skills, and connections. This applies to education in general, not only to medical education.

Knowledge and skills can now be learned far more efficiently with AI. Whether practicing a differential diagnosis on a case or practicing a medical interview, AI plays the other side and gives feedback on the spot. Non-technical skills that are hard to explain in words can also be practiced with scenarios.

Fostering connections, on the other hand, has not yet received enough attention in education. The ability to build relationships with patients, families, people in the community, and other professions is hard to acquire only in classrooms and simulations. The faster AI makes learning knowledge and skills, the more weight the time spent fostering connections carries.

The ironies of automation

In 1983, the psychologist Lisanne Bainbridge wrote a paper called “Ironies of automation” about automating industrial process control1. As automation advances, people are left only with the work that could not be automated. Because machines take over routine work, people lose opportunities to use their skills. Yet it is people who must respond when the machines stop working well. By taking away the easy parts of the task, Bainbridge wrote, automation can make the difficult parts of the human operator’s task more difficult.

Applied to medicine, the more AI drafts differential diagnoses and explanations, the more what remains for clinicians are the situations where AI cannot answer and the situations where they must notice that AI’s answer is wrong. Some training hospitals have residents build clinical reasoning without AI in the early period of training, and let them start using it only after assessing their ability. Noticing where AI has gone wrong requires experience of thinking without AI.

Environments with rules and without

The psychologist Robin Hogarth and colleagues divided environments in which we learn from experience into two kinds2: “kind” environments, where the results of our actions come back quickly and accurately, and “wicked” environments, where feedback is delayed, missing, or misleading. Chess, with fixed rules and immediate results, is close to a kind environment, and it is a field where AI surpassed humans early.

Medicine is a wicked environment. The results of treatment are not known right away, and treatment for one disease affects others. An older patient with many chronic conditions lives alone and is unsure how to take their medicines. Such combinations do not appear in the same form anywhere in a textbook.

Simulation is a way to deliberately create a kind environment inside a wicked reality. Because results come back quickly and you can try again, knowledge and skills deepen reliably. But the feedback that comes back in real practice is delayed and unclear. Practice in taking on decisions there cannot be done fully inside a kind environment.

Learning to take responsibility

In a 1978 report, the US Institute of Medicine (IOM) defined primary care as “accessible, comprehensive, coordinated, and continual care provided by accountable providers of health services”3. Of these, coordination and continuity relate to connections, and the last, accountability, refers to clinicians taking on their own decisions and their results.

In simulation, it is common to tell learners, “It’s fine to fail, so try things freely.” This safety is what lets learners try boldly. However, I think that responsibility grows by taking on small risks, little by little, in places where the results of one’s decisions reach someone. Seeing the same patients over time at a community clinic or hospital, and meeting people in the community and other professions face to face: in such placements, being entrusted with small decisions and seeing their results through fosters a sense of responsibility as a clinician.

AI is useful here too. If you speak about what you saw and thought during a placement, transcribe it, and reflect on it with AI, you can deepen your thinking. Before meeting people in a community, you can also use AI to look up materials on the local health care situation. Increasing the time for building connections is also one of AI’s uses.

The doorman’s job

The advertising executive Rory Sutherland has written about what he calls the “doorman fallacy”4. A hotel defined its doorman’s job as “opening the door” and replaced him with an automatic door. But the doorman had been hailing taxis, recognizing regular guests, keeping the entrance safe, and signaling the hotel’s status to its guests. Because that work lay outside the job written on his name badge, the hotel noticed its loss only after the replacement.

If we define a physician’s job as “assigning a diagnosis,” it looks replaceable by AI. In real practice, more time is spent after the diagnosis is settled. Deciding on treatment in line with the patient’s life and wishes, coordinating with family and other professions, and continuing to take on the results: that is the physician’s job.

What medical education in the age of learning with AI must secure are places to learn the work not written on the name badge. Hone knowledge and skills in places where it is fine to fail, and then go out into the community, connect with people, and take on small responsibilities. I think the task is to provide both, continuously, from undergraduate through postgraduate training.

References

  1. Bainbridge L. Ironies of automation. Automatica. 1983;19(6):775-779. https://doi.org/10.1016/0005-1098(83)90046-8
  2. Hogarth RM, Lejarraga T, Soyer E. The two settings of kind and wicked learning environments. Curr Dir Psychol Sci. 2015;24(5):379-385. https://doi.org/10.1177/0963721415591878
  3. Institute of Medicine. A Manpower Policy for Primary Health Care: Report of a Study. Washington, DC: National Academy of Sciences; 1978. https://nap.nationalacademies.org/read/9932/chapter/3
  4. Sutherland R. Alchemy: The Magic of Original Thinking in a World of Mind-Numbing Conformity. London: WH Allen; 2019.
English translation of the Japanese article.