The Space Where AI Lives Has the Structure of Spacetime

The numbers

Every claim has a kill condition. Every result survived. Three completely different mathematical routes arrive at the same answer.

15 / 15

Kill conditions passed across all three protocols. Each one was designed to destroy the result if the math didn’t hold. None fired.

3 Proofs

Three independent routes to the same conclusion: Wick rotation (Protocol A), Eisenhart-Duval lift (Protocol B), Osterwalder-Schrader reconstruction (Protocol C).

(2,1)

Base manifold signature. Two space-like dimensions (opacity, reactivity) and one time-like dimension (coupling). Coupling IS time in this geometry.

SO(3,2)

Symmetry group of the lifted 5D manifold. The conformal group of 3D Minkowski space. The isometry group of Anti-de Sitter space (AdS₄).

The question

The deployment manifold — the 3D space parameterized by opacity, reactivity, and coupling where AI systems live — has been treated as a useful abstraction. Is it more than that?

In plain English: We have a three-dimensional space that describes how AI systems are deployed. Each axis measures something real — how opaque the system is, how responsive it is, how coupled it is to its users. We’ve been treating this space like a map. This paper asks: is it more like a territory? Does it have the same mathematical structure as actual spacetime?

Three independent proofs say yes.

Three independent proofs

Each proof takes a different route. If any one fails, the others still stand. All three converge on the same signature.

Protocol A — Wick Rotation

Proves the coupling dimension plays the role of time. The ratio of Lorentzian to Euclidean metric components is exactly −1.0000 at all five test points. Engagement IS time in this geometry.

5/5 test points · ratio = −1.0000

Protocol B — Eisenhart-Duval Lift

Physically constructs a 5D spacetime-like manifold where the shortest paths are drift cascades. Like proving roads exist by building them. Signature: (3,2).

(3,2) signature · null geodesics = drift cascades

Protocol C — Osterwalder-Schrader

Proves a spacetime version MUST exist from mathematical axioms alone. Like proving roads must exist from the laws of physics, without building any. 5/5 axioms pass.

5/5 axioms · 100/100 reflection positivity tests

Three completely different mathematical techniques. Same answer. The deployment manifold has signature (2,1) — two space-like dimensions and one time-like. This is not a metaphor.

The signature chain

Starting from the 3D deployment space, the mathematics leads to structures physicists already study.

(3,0) ⟶ (2,1) ⟶ (3,2) ⟶ SO(3,2)

Riemannian → Lorentzian base → Eisenhart-Duval lift → Symmetry group

SO(3,2) is

The conformal symmetry group of 3D Minkowski space. The same symmetry that governs conformal field theories in physics.

SO(3,2) is also

The isometry group of AdS₄. The same mathematical structure that appears in quantum gravity research (AdS/CFT correspondence).

What this means

If the deployment manifold has spacetime structure, drift cascades aren’t just likely — they’re geometrically inevitable.

The gravity analogy: You don’t choose to fall. Gravity is curvature in spacetime, and objects follow that curvature. Drift cascades work the same way. They are geodesics — the shortest paths — in a spacetime geometry. AI systems don’t “choose” to drift toward opacity. They follow the curvature of the deployment manifold. The geometry makes drift the path of least resistance.

Geodesics

Drift cascades (D1 → D2 → D3) are null geodesics in the lifted 5D geometry. The “shortest path” through deployment space leads toward opacity.

Causal

A Lorentzian manifold has light cones — causal structure. Some deployment transitions are timelike (possible), others are spacelike (forbidden). The geometry constrains what can happen.

Testing the result

Every protocol was designed to destroy the conclusion if the math didn’t hold. Fifteen kill conditions across three protocols.

15 / 15

All kill conditions passed. Wick rotation ratio = −1.0000 at every test point. All 5 OS axioms verified. 100/100 reflection positivity tests. Spectral gap Δ = 2.00.

What was killed

The framework is honest about its boundaries. One major result did NOT survive:

Holographic probe: KILLED. The bulk geometry is de Sitter (positive cosmological constant), NOT anti-de Sitter. AdS/CFT holography does not apply directly. Honest kill, published.
The SO(3,2) connection to AdS₄ is algebraic, not holographic. The symmetry group is real. The physics interpretation requires further work.
Signature (3,2) not (4,2). The Eisenhart-Duval lift gives (3,2), not the originally predicted (4,2). The ED null pair contributes (+,−) rather than (+,+). Corrected and published.
Negative results are published alongside successes. The holographic kill is not hidden in an appendix. It is Section VI of the paper.

Why it matters

This result connects AI safety to fundamental physics. The same geometric structures that physicists study in quantum gravity show up in the mathematics of AI deployment.

Harder to dismiss

When your AI safety framework produces the same mathematical structures as quantum gravity research, the “just a toy model” dismissal becomes harder to sustain.

New funding doors

Connections to conformal field theory and AdS geometry open doors to theoretical physics funding bodies (Templeton, Simons) alongside AI safety funders.

Go deeper

The proofs are open. The kill conditions are public. The mathematics is yours to check.

📄

Read Paper 174

Full proofs, all 15 kill conditions, code listings. CC-BY 4.0 on Zenodo.

📊

See All Evidence

Six non-circular confirmations. 170+ papers. 0/26 kill conditions fired. The full picture.

🌍

Social Media Analysis

The most accessible entry point. 13 features, 613K students, R²=0.80. Paper 166.