Nature Needs Mathematics for Its Rigorous Description But is Computation at the Root of Nature

Feynman’s made a lecture and book that mathematics is needed to describe nature. It (in The Relation of Mathematics to Physics) is descriptive and he is slightly awestruck. The fundamental laws turn out to be statable in mathematics, and there’s no obvious reason they had to be. It’s the same puzzle Wigner named the “unreasonable effectiveness of mathematics.” Crucially, Feynman doesn’t claim nature is mathematics — only that math is the language in which its regularities become sayable.

Google Deepmind Demis Hassabis pushes further. His claim is that most natural systems can be simulated or learned—reverse engineered—by a classical system, and that this would support a conjecture that information is the most fundamental unit of physics, more so than matter and energy. The through-line is Levinthal’s paradox. A protein has astronomically many configurations yet folds in milliseconds, so AlphaFold’s ability to predict the answer suggests reality occupies a “tiny, learnable corner” of possibility space. He’s careful to separate this from Bostrom — no simulator, no parent reality, just a universe whose fabric is informational.

Hassanis proposes an empirical research program. IF learning systems keep compressing whatever nature produces, that’s accumulating evidence for information-theoretic order — and if they hit hard walls, that’s evidence too.

Protein Folding was never actually shown to be uncomputable — Levinthal’s paradox is about naive brute-force enumeration, not about the physical process. Nature doesn’t search the configuration space; it slides down an energy gradient. Hassabis half-concedes this himself when he says there must be some landscape in the energy landscape, some gradient you can follow. But once you grant a smooth guiding landscape, fast folding stops being paradoxical, and AlphaFold learning that landscape is impressive engineering rather than evidence that reality is “made of computation.”

“Learnable” and “computational at root” aren’t the same claim. Evolved proteins are learnable partly because evolution and physics select for stable, low-entropy, regular structures — the data distribution is pre-filtered to be compressible. That’s a fact about which structures survive, not necessarily about the ontology underneath them. This is exactly the standard view Hassabis is arguing against (information as a description we lay on top), and AlphaFold is consistent with both readings. It doesn’t discriminate between them.

Hassabis is making a conjecture. It’s a testable restatement of the Feynman–Wigner puzzle rather than new proof of a computational universe. AlphaFold is evidence that a large swath of evolved biological structure is compressible and learnable — which is remarkable and useful. The leap from “compressible” to “information is more fundamental than matter and energy” is a philosophical interpretation the folding result is compatible with but does not compel. Hassabis mostly says this himself and promises an actual paper. The real test of the idea is whether it can specify in advance where the walls should be and then be caught out if nature falls on the wrong side.

Stephen Wolfram’s Computational Universe (Hypergraph Rewriting / Physics Project)

Stephen Wolfram says Physics doesn’t explain the universe. Computation does.

Wolfram has the a developed program. He argues the universe emerges from simple discrete computational rules (hypergraph rewriting) applied repeatedly, producing space, time, particles, relativity, quantum mechanics, and the Second Law via computational irreducibility (no shortcuts for many processes; time is the “doing” of irreducible computation). The “ruliad” is the entangled limit of all possible computations. Our physics is a slice perceived by observers like us.

What it would take to prove/validate (per Wolfram and his project).

Identify the specific simple underlying rule(s) that, when run, reproduce all known physics (GR, QM, Standard Model particles, cosmology, etc.) with minimal or no free parameters/tweaking.
Derive existing laws as emergent “reducible” pockets within irreducible computation.
Make novel, testable predictions (e.g., about quantum gravity, black holes, or early universe) that match experiments/observations.
Confirm computational universality and observer-dependent effects (how we sample the ruliad shapes perceived laws).

This is an active research program with tools for enumeration and matching. No external “computer” or simulator is needed—the rules are the computation.

Digital Physics, Lattice/Discrete Spacetime Tests

Pioneers like Konrad Zuse and Edward Fredkin viewed the universe as a cellular automaton or discrete computation. Modern extensions look for Planck-scale (or coarser) discreteness.

Proposed signatures/tests

Lattice artifacts – Anisotropies, cutoffs, or dispersion relation violations in high-energy cosmic rays, particle spectra, or neutrino propagation (from discrete spacetime “grid” effects in simulations).

Interferometry – Matter interferometers (neutrons) detecting phase shifts or modified dispersion from discrete structure—potentially sensitive to very small scales.

Other observables – Deviations in CMB, gravitational waves, or Lorentz invariance at extreme energies. Changing fundamental constants (John Barrow’s idea of simulation “fixes”).

Papers outline mathematical criteria for detectability and required experimental sensitivity.

Progress coudld come from combining theory (rule-finding or new laws like infodynamics) with precision experiments probing fundamental scales or information signatures. No definitive proof exists yet, but there are frameworks and proposals and various experiments.