arXiv:2611.03596
Spacetime Emergence in the Memory of a Learning Agent
Philip A. LeMaitre, T. Rick Perche
Abstract
We show that the causal structure of spacetime can emerge within the memory of a learning agent performing a quantum information task. We train a model-based reinforcement-learning agent to maximize the quantum mutual information between localized quantum probes interacting with a quantum field in Minkowski spacetime, without directly providing it with information regarding the spacetime metric or causal structure. In learning to accomplish this task, the reinforcement-learning agent constructs an internal predictive model whose planner develops the characteristic null structure of spacetime, allowing us to reconstruct the causal structure of the Minkowski metric with high accuracy. Within this task, spacetime geometry is obtained from an agent's successful predictions and actions, providing a first step toward showing that spacetime concepts emerge from operational quantum experience.
AI Summary
Can an agent trained on operational quantum-information data acquire an internal representation from which spacetime causal structure can be recovered? This paper studies a model-based reinforcement-learning agent navigating detector placements in a simulated quantum-field environment, then reconstructs null directions from its learned reward predictions. Two gapless, Gaussian-smeared Unruh–DeWitt qubits interact with a massless scalar field in the Minkowski vacuum. One detector is fixed; the other occupies a $301\times301$ grid in an $(x,t)$ slice. The agent receives normalized placement coordinates and rewards based on a fourth-root transform of detector mutual information, with progress bonuses and penalties. An encoder–decoder, latent transition model, and reward model support one-step action selection. Five independently trained agents each reach the coincident-detector goal in all 1,000 evaluation episodes, averaging approximately 104 steps. For reconstruction, action-score fields are averaged across agents, converted into a scalar field, and processed using a local ridge mask and branch-selection procedure. A quadratic null-condition fit, fixing $g_{xx}=1$, gives $g_{tt}=-0.999878$, $g_{tx}=7.787\times10^{-8}$, and null slopes approximately $\pm0.999939$. These are fitted values, not uncertainty estimates. The stated limitations are that geometry is recovered through externally supplied geometric post-processing, not explicitly inferred by the agent; coordinates are supplied; and mutual information mixes causal and state-dependent correlations. The demonstration concerns straight null branches in a flat two-dimensional slice. Coordinate-change and detector-smearing robustness, curved and higher-dimensional settings, and tasks requiring geometric understanding remain untested. Only conformal structure is reconstructed.
Error found: The abstract leaves the dimensional scope unspecified, whereas the demonstrated reconstruction recovers only the conformal null geometry of a fixed (1+1)-dimensional slice rather than the full (3+1)-dimensional Minkowski causal structure.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 19 pages, 7 figures