Pattern Extraction vs. Understanding

When AlphaFold predicts a protein's three-dimensional structure from its amino acid sequence with accuracy rivaling crystallographic experiment, it is performing Rosalind Franklin's core operation: recovering the architecture of the invisible from the record of how it bends the things you can see. The continuity between Franklin's basement at King's College London in 1952 and the data centers running protein-folding models today is not metaphorical—it is the same fundamental inverse problem, now automated at the scale of the proteome. But the echo exposes a gap that is the whole question: when the machine recovers the structure, has it done what Franklin did, or only the output half of it? Franklin's inference was disciplined by an understanding of physics. She knew why the spots fell where they fell on the diffraction film; she could reason about what each feature of the pattern implied and what it did not; she could distinguish a real signal from an artifact of how