The second mechanic didn’t know more about engines. He stopped looking at the engine.
A car won’t run. The mechanic has worked through every diagnostic procedure in the manual for the engine, and nothing turns up. A second mechanic arrives, listens for thirty seconds, and says: “You’re diagnosing the wrong system. The problem is electrical, not mechanical.”
What did the second mechanic have that the first didn’t? Not better methods: the first mechanic’s procedures were correct. Not necessarily more knowledge either. People who step into messy, volatile situations and turn them into routine problems often have less relevant factual knowledge than the people who found the situation impossible, and no special skill in applying technical methods. What they have is the ability to operate one level above the methods. Once they change which problem is being solved, textbook methods that were failing to come to grips with anomalies start working again. David Chapman calls this meta-rationality.
Changing the description
The first mechanic was not bad at mechanics. He was solving the problem as described, “something is wrong with the engine,” and every refinement of method made him better at the wrong task. More sophistication in the wrong frame produces only more sophisticated failure. That is why the fix looks like magic from inside: the problem being solved was not the problem that mattered.
So meta-rationality is the skill of changing the description of the problem, which is categorically different from the skill of solving the problem as described. In practice it shows up as small moves. You notice a factor others overlooked, not because you are more observant but because you are not committed to the current framing. You point out a gap between theory and reality, which is where, as Weinberg argues, problem definitions usually go wrong. You ask whether a difficulty everyone is struggling with can simply be ignored. You borrow a concept from a distant field. Each move changes the description, and a new description makes different methods relevant. Most education trains the second skill. Almost none trains the first.
Frames are cheap
It would be easy to conclude that the skill is generating alternative framings. Mostly it isn’t. LLMs make candidate frames cheap: the crystal at the frontier can restate one anomaly through five disciplines in seconds. The meta-rational act remains human: noticing which description exposes a testable assumption and which four merely sound clever. The second mechanic did not list every system in the car. He named one he could check.
So explore problem-space before solution-space. The person who spends an hour reframing the problem before touching a method will outperform the person who spends that hour applying the wrong method with increasing desperation. What stops most people is that reframing feels like wasting time, while running the next diagnostic feels like work. The first mechanic was never idle. That was the trouble.
References:
- David Chapman, Meta-Rationality
- Morgan Housel, Useful and Overlooked Skills