Member-only story
The Jump Before the Method
Sensemaking, abduction, and why AI makes expertise more necessary — not less
Ask what makes science scientific, and the usual answer is the scientific method. Observe. Form a hypothesis. Derive predictions. Run an experiment. Analyse the results. Revise the hypothesis.
This is a useful discipline. It is also an incomplete account of discovery.
The sequence begins only after someone has already decided what deserves attention, what kind of phenomenon is occurring, which concepts are relevant, and what hypothesis is worth testing. By the time a clean problem statement appears, much of the most consequential intellectual work has already happened.
A recent Google DeepMind position paper, provocatively titled LLMs Can’t Jump, gives this missing work a name. Drawing on a diagram Albert Einstein once sent to Maurice Solovine, the paper describes scientific invention as a movement from sense experience, E, to a system of axioms, A, by way of an intuitive conceptual Jump, J. Logical deduction can unfold the consequences of the axioms, but it cannot tell us which axioms to invent. The paper argues that contemporary AI is increasingly capable of induction and deduction, yet still lacks the abductive mechanism required to make this jump.
