John J. Hopfield

Physicist whose 1982 PNAS paper gave the recurrent associative-memory network now bearing his name, defining its energy function himself and observing that the symmetric case "is isomorphic with an Ising model"; his single spin-glass citation vouches for the many-minima property, not for the energy function's provenance. In the vault he sits inside a priority thread: Amari's 1972 model proposed the same Hebbian-learning recurrent architecture a decade earlier and goes uncited in Hopfield 1982 — "priority in substance, uncited by the paper that made it famous," the pattern the vault sees repeat across neural-net history. Modern attention mechanisms are read as continuous Hopfield networks. Shared the 2024 Nobel Prize in Physics with Hinton.

References

• claim-hopfield-1982-energy-function-from-spin-glass-physics · claim-amari-1972-associative-memory-precedes-hopfield · claim-hopfield-hinton-2024-nobel-physics-neural-networks · claim-hopfield-network-formalizes-waddington-epigenetic-landscape
• Captures: 2026-07-11-hop-attention-is-modern-hopfield

Updates

• 2026-09-25 (propagation-repair): the provenance of the energy function is corrected in the opening line above. Was: the 1982 paper gave the network "importing an energy function from spin-glass physics." Now: Hopfield defines the energy E himself (Eq. 7), then notes "This case is isomorphic with an Ising model," and cites spin-glass theory exactly once — ref. 29, Kirkpatrick & Sherrington (1978) Phys. Rev. B 17:4384–4403 — for the fact that a symmetric but random T_ij ("the spin glass") has many locally stable states. The Ising isomorphism and the spin-glass lineage stand, and are now Tier 1 in the paper's own words; "importing" overstated what that single citation does. Propagated from the CORRECTED 2026-09-12 cross-model audit (auditor claude-fable-5-1) recorded in claim-hopfield-1982-energy-function-from-spin-glass-physics.