AI-Discovered Laws of Mind
Type: intersection (second-order) Slug: intersection—AI-discovered-laws-of-mind Parents: intersection—density-functionals-learnable-nature, intersection—mathematics-of-construction Last updated: 2026-05-14 Epistemic status: Extrapolative
The convergence
Density functionals prove AI can discover correctness properties that human theories missed (intersection H). Mathematics×Construction proposes AI-guided formalisation of the construction system (intersection 10). Combined: the strongest version of the mathematics programme — AI doesn’t just formalise construction; it discovers the governing laws, including correctness properties invisible to human theorists.
The density functional paper is the proof of concept: if AI can find piecewise linearity in quantum chemistry, it can find analogous invariances in cognitive architecture. The mathematics intersection provides the target domain: construction. Together: apply the “correctness property discovery” method from density functionals to neural/behavioural data about construction.
Why this isn’t obvious from either parent
Density Functionals×Learnable Nature (H) identifies the proof of concept but doesn’t specify a target domain beyond quantum chemistry. Mathematics×Construction (10) specifies the target domain (construction) but doesn’t have a method for discovering laws — it proposes AI-guided formalisation, which assumes the laws are already approximately known. Together: the density functional method (train multiple architectures, find convergent properties) can be applied to construction data to discover the laws, not just formalise ones humans already suspect.
Generative prediction
Train multiple neural architectures to perform construction-like tasks (scene completion from partial elements, imagination from stored components). Analyse their learned representations for properties that converge across architectures — analogous to how different density functional networks converge on piecewise linearity. If convergent properties are found, they are candidate “laws of experiential coherence” — discovered rather than assumed.
Falsification: If no convergent correctness properties emerge across different cognitive architectures trained on construction tasks, the law-discovery claim is false.