The basis of expert intuition

There are often stories told of “master clinicians” – individuals who have had several decades of clinical experience who can walk into a room, and after laying eyes on the patient, immediately know (for example) from their temporal wasting and hypophonic voice that they are suffering from amyotrophic lateral sclerosis. In these cases, has the physician developed a unique framework for thinking about the physical exam signs, symptoms, and laboratory values? Or have they had so many experiences with similar patients that they have developed an “intuition” about this type of patient? What is the cognitive process by which a novice’s (often slow and careful) approach to a problem like this transforms into a fast, seemingly intuition-driven approach of an expert? What value does the intuition-driven approach provide? In this post, I explore the differences between critical “principles-based” reasoning versus pattern-matching, arguing that novices often rely on the former while experts often rely on the latter. I then discuss a Bayesian framework to unify principles-based reasoning with intuition and explore the value of using both approaches in reasoning.

If you have ever taken a behavioral psychology class you have probably heard about “System 1” and “System 2” thinking. In this framework, System 1 is a relatively effortless, intuition-based, and fast approach to addressing problems while System 2 is effortful and relies on critical thinking and reasoning to slowly address more challenging questions. This is a version of Dual Process Theory originally proposed by William James and then popularized by Daniel Kahneman in his book Thinking Fast and Slow in 2011. Of course this is merely a framework – many foundational studies in the field have failed to replicate [1] and the molecular basis of thought, decision-making, and behavior is far from well defined – but it remains a useful (albeit controversial) one. This framework is frequently used in medical education, with both systems being useful in clinical decision-making. System 1 is often derided as being more bias-prone and we are encouraged to use System 2 – more effortful, deliberate, reasoning-based approaches (e.g., checklists, frameworks) – to avoid these pitfalls. I argue that “System 1” thinking, when done properly, may actually represent a crucial part of the maturation of a practitioner in a field from novice to expert.

So what is “expert intuition”? To paraphrase Herbert Simon, intuition arises from rapid pattern matching to an extensive library of past experiences. In other words, expert intuition is similar to System 1 thinking above. As described nicely elsewhere[2], this idea is illustrated by observations made with chess grandmasters who take such little time to make excellent moves that it is not feasible for it to be a product of a “careful analysis” (i.e., it is unlikely to be driven by System 2 thinking).

Having established that expert intuition is closer to System 1, how does one learn a new skill at an expert level? It takes an immense amount of deliberate practice to develop the depth of experience required to have expert-level intuition under Simon’s framework. This practice results in the “crystallization” of System 2 thinking. To illustrate, here are three examples:

1. Recall what it was like to first drive a car. Initially, an active cognitive load is required to think about which pedal one’s foot is on, how much to press, when to start braking prior to turning, etc. Driving down a road involves constantly worrying about if your car is in the center of the lane. Eventually one is able to learn to become comfortable enough to drive almost unconsciously, achieving a degree of “expertise” allowing one to drive on a frequently travelled route without much careful thought.
2. Imagine learning to play competitive chess for the first time. A useful approach would be to spend time thinking through core principles and frameworks for how and when to use certain openers, strategies for the mid-game, etc. Eventually, one may no longer need to think through a giant list of opening moves – with enough experience, an intuition develops, and the need for using executive control to flip through frameworks is superseded by a combination of drawing from experiences and an internalized version of those frameworks.
3. Returning to medicine, we are told that one of the most important skills to gain early in residency is to distinguish “sick” from “not sick”. Initially, trainees may explicitly identify specific signs and symptoms that can aid in making this determination. However, with enough experience, this becomes a gut feeling that automatically kicks in when observing any patient.

In these diverse examples, repeated experiences allow us to iterate on our frameworks, internalize the aspects that are actually useful for the task at hand, and begin to subconsciously process sensory input to feed these frameworks without much additional cognitive load. Eventually, we can abstract individual tasks that were once challenging (look in the side-view mirror, check if there is traffic in the adjacent lane, turn the steering wheel, check if the car is centered in the adjacent lane) into a single more complex task (switch to the next lane) which then can be abstracted into an even more complex task (drive home). Instead of assessing each individual piece on a chessboard as a novice might, an expert can see a chessboard and evaluate the entire field at once.

The development of expert intuition does not preclude the use of “careful analysis”. Empirically, when we enter new or unfamiliar situations we can no longer rely on cognitively light heuristics and have to return to intensive reasoning. For example, when someone comfortable with driving enters a complex scenario (e.g., merging onto a fast-moving highway during sudden downpour in the dark), they may go silent as they have to use more cognitive effort to carefully analyze and navigate the situation. In other words, internalized expert intuition is only effective when new scenarios are “within-sample”. For instance, if a clinician has seen several patients with volume overload, their intuitions about treatment will apply well to a new patient with volume overload but not to a patient with a skin infection. There is even a term for the act of stepping outside the bounds of one’s immediate expertise: epistemic trespassing ¹. In medical education, this risk of unintentional bias from relying on instinct over unbiased principles-based reasoning is why “System 2” thinking is often encouraged ².

We can consider a Bayesian framework to try to connect the concepts of expert intuition and “careful analysis”. For a given decision landscape, the expert ability to rapidly draw from prior experience can be thought of as a prior. The System 2-style framework-based careful analysis can be thought of as a transformation of the current data. The product of the two is the posterior – the set of probabilities one could use to make a decision. If the current data is limited – for instance, a match of speed chess where there is little time to think, or an urgent clinical situation where time or information is limited – then one has to lean particularly heavily on the prior, i.e., relying on expert intuition arising from past experience. On the contrary, in novel situations, one may need to rely more on the available data than one’s prior experiences. Going back to the example of driving, even though one may be able to navigate typical routes without much thought, a novel situation (say, a complex roundabout or heavy downpour) may require a return to using active cognition as one processes the new situation.

Clinical teams, who often approach simple cases differently from atypical or “medical mystery” presentations, offer an interesting case study. Clinical teams in teaching hospitals often span many levels of expertise, including medical students, first-year residents, senior residents, and attendings. As a junior trainee it can sometimes be hard to follow the team’s clinical decision making as so much is reliant on prior experience and “System 1” expert reasoning. Different teams will often implement different clinical decisions despite having the same objective data as people have differing priors – diuretic dosing, for instance, is often attributed to “style”. In contrast, when the diagnosis is more uncertain and there is little past experience to rely on, teams will often revert to a “System 2” style thinking. This entails a careful and very explicit systems-based differential diagnosis with subsequent diagnostics and treatments for each possible diagnosis. Counterintuitively, it is in this setting that junior trainees can often contribute more – though they may lack a large corpus of past experience, they have the foundational knowledge to consider, in a principled way, a broad differential of possible drivers of disease based on symptoms and laboratory values. This could also arise if a seemingly “typical” presentation begins to behave atypically, requiring a reset by the team (for instance, a UTI that does not improve with basic antibiotics, prompting a “broadening” of one’s thinking to include a deeper infection, resistant organisms, autoimmune causes, and more). Strong medical training programs encourage such a principled approach for even the simplest presentations so as to make such reasoning second nature.

In thinking through the basis of expert intuition, I come away with a simple conclusion: one must mindfully leverage both styles of thinking depending on the situation one finds themselves in. In contrast to common sentiment which derides System 1 thinking, in situations where there is little objective information to rely on and there is ample prior experience, relying on heuristics gleaned from this past experience is the “special sauce” that only an expert can have. However, it is important to critically assess how well past experience fits the situation at hand. One must be willing to switch back to basic frameworks and “careful analysis” to appropriately reason through novel situations.

References