The Problem Solver: Educating to Critical Thinking

Judgment concentrates where the model runs out, ethical reasoning is the connective tissue binding the whole of the human advantage, and evaluating truth is a discipline of intellectual honesty more than a technique. Each of those capacities is exercised under uncertainty, and none can be exercised by a physician who cannot analyze a problem in the first place. In an age of machine intelligence and avalanches of data, critical thinking emerges as the decisive ability, and current training assumes it far more reliably than it teaches it.

Problem solving has a structure that can be taught, and two bodies of work supply it.

Zygmunt Pizlo's Problem Solving: Cognitive Mechanisms and Formal Models integrates behavioral research with formal computational models, and several of its distinctions transfer directly to clinical reasoning. It separates search problems, where the solution space is defined and the task is efficient traversal, from insight problems, where progress requires reconceiving the problem itself.

It shows how human beings cope with exponentially large search spaces through hierarchical representations that approximate instead of enumerating, and through heuristics organized around a sense of distance and direction toward a goal. It connects perception to problem solving, treating both as inference over ill-posed problems where the data underdetermine the answer. And it locates scientific discovery itself as a species of insightful problem solving.⁶¹

The decisive claim concerns mental representations: how a problem is represented largely determines whether it can be solved at all. A problem well represented is often nearly solved. A problem badly represented is frequently insoluble by any amount of effort or data. This reframes what a diagnostic error usually is. It is less often a failure of knowledge retrieval than a failure of representation, in which the clinician solved, efficiently and correctly, a problem that was not the one in front of them.

The complementary source is older. George Pólya's How to Solve It (1945) supplies a four-phase heuristic that has survived eighty years because it describes what expert problem solvers actually do: understand the problem; devise a plan, finding the connection between the data and the unknown, and considering auxiliary problems when no immediate connection appears; carry out the plan, checking each step; and look back, examining the solution obtained.

Pólya's power lies in the questions beneath each phase. What is the unknown? What are the data? What is the condition? Is the condition sufficient? Have you seen this before? Do you know a related problem? Could you restate it? Did you use all the data? Can you check the result? Can you derive it differently? Can you use the result, or the method, for some other problem?

These questions are conspicuously absent from clinical training. Did you use all the data is a ward-round question. Could you restate the problem is exactly the representational move Pizlo identifies as decisive. Can you derive the result differently is the discipline that would catch a confident anchoring error. Pólya's fourth phase, looking back, is the one medicine most systematically skips, since the case moves on and the reasoning is rarely revisited when the outcome is known. In an era when a machine will supply a fluent answer to phases two and three, the physician's distinctive contribution concentrates in the first and the fourth: formulating the problem correctly and examining critically what has been produced.

Medicine has historically trained the search case well and the insight case hardly at all. Differential diagnosis is a search procedure over a defined space, taught explicitly and examined relentlessly. But the cases that will remain with physicians as machines absorb the routine are exactly the insight cases, where the presentation does not fit, the data conflict, and the represented problem is the wrong problem.

Teaching the insight case is possible, and two moves would carry most of it. The first is to make re-representation a required step rather than a talent, by building into case discussion an explicit pause before commitment in which the trainee states the problem a second way: as a different organ system; as a different time course; as a social problem presenting physiologically; or as the patient's problem rather than that of the team. Stating an alternative representation survives being drilled, and it is the working half of Pólya's first phase.

The second is to build case libraries around representation failures rather than knowledge failures. Morbidity and mortality conferences overwhelmingly present cases in which something was not known; the instructive cases are those in which everything was known and the problem was framed wrongly.

Cultivating critical thinking must therefore become a central aspect of medical education throughout the continuum, and the content of that instruction can be specified. It means explicit teaching of evidence appraisal and causal inference, and of cognitive bias recognition under time pressure, which is the condition under which biases actually operate and under which they are almost never practiced. It means generating and testing alternative problem representations, the Pizlo move made into a habit, not a talent, and seeking disconfirming evidence as routine instead of as virtue. And it means distinguishing a well-formed question from a poorly formed one, which is increasingly the highest-value skill in a world where any well-formed question will be answered instantly by a machine.

Two pedagogical implications follow that current training largely ignores. The first is that critical thinking must be practiced on one's own reasoning and not only on the literature. Journal club teaches the appraisal of others' work; almost nothing in the curriculum teaches the appraisal of one's own diagnostic process, and the simulation debrief described in the next chapter is the only reliable instrument medicine has for it.

The second is that the discipline requires exposure to being wrong in a setting where being wrong is survivable, which again points to simulation, and which explains why a culture of individual assessment is so corrosive to the very capacity it claims to measure. A trainee who has learned that error is career-threatening will not surface the reasoning that produced it, and reasoning that is not surfaced cannot be corrected.

It need not be confined to medical issues. Critical thinking extends to other dimensions and issues in the world, and there is a case, pedagogical and not political, for practicing it on material where the trainee has no professional stake, since it is easier to notice a flawed argument when one is indifferent to its conclusion.

This carries a call for dialogue and for the apolitical, subjecting all viewpoints to scrutiny. A training environment in which certain conclusions are protected from examination does not teach critical thinking. It teaches which conclusions are protected. The discipline is content-neutral, or it is not the discipline.