Hacker News

"Intelligence Without Representation: How a Dynamical System Discovered the Laws of Logic Simply by Trying to Restore Symmetry"

---

This paper presents a radical proposition: what if intelligence isn't about storing knowledge, but about continuously restoring broken symmetries?

The authors build a mathematical framework where an agent has no memory, no representations, no goals—only the ability to measure how much the current flow of experience violates its learned symmetries, and act to reduce those violations. Memory, prediction, and reasoning emerge naturally from these dynamics.

Then comes the astonishing part: they feed this system nothing but 16 basic logical axioms encoded as symbol frequencies, and it discovers the fundamental structure of propositional logic on its own. Six learned generators emerge—corresponding exactly to variable negation, De Morgan duality, commutativity, and distribution—achieving near-perfect reconstruction of logical relationships (0.015 discrepancy vs. 0.19 for cyclic baselines). When they compose these generators, the system produces novel valid tautologies not present in the training data.

The system never represented "De Morgan's law" as a proposition. Yet it could generate De Morgan transformations because the dynamical flow had carved out a direction in its state space corresponding to that operation. It didn't learn about logic—it became logical.

The implications are profound: perhaps knowledge isn't something we have, but something we do—a continuous enactment of symmetry with the world.

This is AI generated bullshit.