CONCEPT
Form as Diagram of Forces
D'Arcy Thompson's conviction that every shape is a frozen record of the forces that made it—and that from the form, the forces can be read back—extended by the cycle into a research programme for the interpretability of trained neural networks.
"The form of an object," D'Arcy Thompson wrote in
On Growth and Form, "is a 'diagram of forces,' in this sense, at least, that from it we can judge of or deduce the forces that are acting or have acted upon it." The shell's logarithmic spiral encodes the mathematics of its growth; the bone's trabeculae encode the stresses it bore; the hexagonal honeycomb cell encodes the surface tension that packed it. In every case the shape is not arbitrary but
legible—a compressed record of causal history that a sufficiently prepared observer can decode. The
[YOU] on AI cycle extends this claim into the age of
neural networks: a trained model is a diagram of the data, the objective function, the architectural constraints, and the optimization dynamics that produced it, and the enterprise of
interpretability is the enterprise of reading those forces back out of the form. The parallel