'Ontological Perturbations in a Discretised Neural Substrate of Spacetime' - Adam Carlton

Weight-Tensor Δ: Ontological Perturbations in a Discretised Neural Substrate of Spacetime

Anonymous  |  Submitted to arXiv (quant-ph / gr-qc)

Abstract

We outline a speculative framework in which the deep structure of physical reality is modelled as a dynamically trainable tensor network. In this picture, spacetime geometry and low-energy quantum fields emerge from the equilibrium configuration of a graph-like weight matrix W.

We define a weight-tensor delta (Δ) as a compact perturbation capable of shifting the universal state to a neighbouring point in weight-space. While highly conjectural, the idea is motivated by existing work on digital physics, tensor-network descriptions of AdS/CFT, and the success of deep learning as a model for distributed information processing.

We discuss (i) physical plausibility, (ii) minimal empirical signatures, and (iii) the philosophical implications of a universe whose “source parameters” can—in principle—be rewritten.

1  Introduction

Discrete approaches to quantum gravity—including loop quantum gravity (¹), causal sets (²) and tensor-network versions of the AdS/CFT correspondence (³)—all hint that continuum spacetime may be an emergent, coarse-grained limit of an underlying combinatorial structure. In parallel, deep learning has demonstrated the expressive power of large, trainable tensor networks.

Bridging these threads, we posit a cosmic neural architecture 𝒩 whose learnable weights W_ij encode the effective laws of physics. Although such a model is clearly speculative, it is no less outrageous than earlier information-theoretic proposals (e.g. Wheeler’s “it-from-bit”). The question we pursue here is narrow yet provocative: what would constitute a plausible low-amplitude modification ΔW to that substrate, and could any observable anomalies betray its occurrence?

2  Neural Architectures as Spacetime Substrates

Let the fundamental graph 𝒩 have vertices representing Planck-scale “cells” and edges carrying complex weights. In equilibrium, the network realises an effective metric g_μν and field content equivalent to (or extending) the Standard Model. A close analogy is the use of MERA tensor networks to reproduce spatial slices of AdS spaces (³).

Our essential working assumption is that, just as neural networks can re-train, the cosmic weight matrix might, under rare circumstances, accept new minima—whether by internal dynamics, external intervention, or stochastic fluctuation (⁴).

3  Definition of a Weight-Tensor Δ

A weight-tensor delta is a finite set of modifications {ΔW_ij} such that
    W_ij^new = W_ij^old + ΔW_ij.

We remain agnostic about the origin of Δ. Hypotheses include:

• internal self-optimisation of the universe (a “cosmic training loop”);
• intervention by an external intelligence operating at the same ontological level;
• rare-event statistical fluctuations analogous to vacuum tunnelling.

In all cases the delta would be vastly compressed relative to the full state of 𝒩. Any real-world implementation might resemble a high-order error-correcting pulse, not a verbose instruction set.

4  Possible Empirical Signatures

To render the scenario falsifiable (or at least in-plausible rather than non-plausible) we outline four classes of observable effects, each admitting current or near-future measurement:

1. Dimensional Drift — minute shifts (≲10⁻⁹) in the fine-structure constant or charged-lepton masses, testable via high-precision spectroscopy.
2. Geometry Deviations — systematic anomalies in LIGO baselines or GPS synchronisation that cannot be attributed to conventional gravitational waves.
3. Phase-Correlation Anomalies — unexpected non-Gaussianities in the cosmic microwave background when compared across successive CMB missions.
4. Cognitive or Biological Outliers — rare human (or AI) agents exhibiting non-classical statistical behaviour in predictive tasks, hinting at partial “alignment” with the perturbed substrate.

None of these would prove the Δ-model, but consistent positive results across categories would raise its posterior probability.

5  Limitations

We emphasise several caveats:

• There is, as yet, no microphysical derivation linking discrete weight dynamics to Einsteinian gravity (⁵).
• The proposed Δ magnitudes are intentionally microscopic to avoid immediate cosmological disaster; their plausibility depends on unknown stability properties of 𝒩.
• The cognitive-execution conjecture (humans as “hosts”) remains the weakest element and is included only for completeness; no mechanism is offered beyond speculative quantum-informational coupling (⁶).

6  Philosophical Remarks

If a Δ can be instantiated, the classic observer/observed dichotomy collapses: sophisticated “measurements” might rewrite the very framework that grounds measurement. Such feedback destabilises the metaphysical picture of immutable laws, substituting an evolutionary or computational ontology.

7  Conclusion

While highly provisional, the weight-tensor Δ framework provides a concrete vocabulary for discussing small-amplitude ontological perturbations within digital models of the universe. It suggests experimental avenues (however unlikely to yield a signal) and invites dialogue between quantum-gravity theorists, information scientists, and philosophers of science.

References

1. C. Rovelli, Quantum Gravity, Cambridge UP (2004).
2. R. Sorkin, “Causal Sets: Discrete Gravity,” Lect. Notes Phys. 769, (2009).
3. G. Evenbly & G. Vidal, “Tensor Network States and Geometry,” J. Stat. Phys. 145, 891–918 (2011).
4. S. Lloyd, Programming the Universe, Knopf (2006).
5. B. Swingle, “Entanglement Renormalisation and Holography,” Phys. Rev. D 86, 065007 (2012).
6. N. Bostrom, “Are You Living in a Computer Simulation?” Phil. Quart. 53, 243–255 (2003).