Clinical reasoning is done by the whole team and its tools
Distributed cognition treats thinking as something that happens across people, artefacts, and the settings they work in. A ward round reasons with its charts, its handover sheet, and everyone in the room. AI joins that system as another participant, and the clinician’s work becomes shaping how well the whole system thinks.
Distributed cognition
One-sentence definition: Distributed cognition is the view that cognitive work is carried out by systems of people, tools, and environments, and that those systems have thinking properties no individual in them has alone.
Where the idea comes from
Edwin Hutchins (1995) developed it by studying the navigation team on the bridge of a US Navy ship. Bringing a large vessel into port is a cognitive task, but no single person on the bridge performs it. Sailors on the wings of the bridge take bearings on landmarks and call them out, a recorder logs them, and a plotter uses purpose-built instruments to mark them on the chart and fix the ship’s position. The computation happens in the coordination between them and their tools, and nobody holds all of it. Hutchins’s point is that the system has cognitive properties that differ from the properties of the people in it, and that “humans create their cognitive powers by creating the environments in which they exercise those powers”.
Clark and Chalmers (1998) made a related argument about individuals, which they called the extended mind: when a notebook reliably does the work that memory would otherwise do, it’s as much a part of the person’s thinking as memory is.
Distributed cognition in healthcare
Clinical reasoning already works this way. A nurse on a surgical ward reasons with the observation chart, the early warning score, the drug chart, the pharmacist’s note, the previous shift’s handover, and the patient’s own account of how they feel. Take away the chart and the same nurse reasons less well. Much of what we call an individual clinician’s judgement is the product of a system that has been built around them, including guidelines, checklists, and the colleagues they can ask.
Coiera (2020) argues that AI changes the make-up of this system. From a distributed cognition perspective, decisions in healthcare organisations “emerge not from isolated internal processes inside the brains of humans or the code of machines, but from the many distributed interactions that occur between these actors”, and AI is now one of those actors, with growing autonomy. Patients are bringing their own AI into it too, having researched their condition and prepared their questions before the appointment. A consultation can now involve two people and two AI systems, each carrying different context.
Much of the debate about AI in healthcare compares a clinician with an algorithm on a single task. From this perspective, the question worth asking is how the system should be arranged. The clinician decides what the AI sees, what it’s asked to do, what gets checked and by whom, and where their own judgement sits. Coiera puts the consequence plainly: “when cognition is distributed between human and computational agents, so is agency and accountability”. Arranging the system well includes knowing where it fails. Lyell and Coiera (2017) reviewed the evidence on automation bias, where people over-rely on decision support and stop checking, and found it even when people were doing a single task, typically a diagnostic one where the output was hard to verify.
For education, this means students need practice reasoning inside systems like these (with records, colleagues, patients, and AI) as well as practice reasoning alone. A simulated ward shift, where a student has to pick up a handover, read the charts, check a dose with the pharmacist, and decide what to escalate, tests something a written exam can’t reach. Assessments that test only what a student can recall unaided measure one part of a capability that, in practice, is always exercised with others.
Sources
- Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
- Coiera, E. (2020). The cognitive health system. The Lancet, 395(10222), 463–466. https://doi.org/10.1016/S0140-6736(19)32987-3
- Hutchins, E. (1995). Cognition in the wild. MIT Press. https://doi.org/10.7551/mitpress/1881.001.0001
- Lyell, D., & Coiera, E. (2017). Automation bias and verification complexity: A systematic review. Journal of the American Medical Informatics Association, 24(2), 423–431. https://doi.org/10.1093/jamia/ocw105