This paper questions the comprehension limits of current AI paradigms and emerging world models by asking: how fast can reality change before an intelligent system fails to perceive, reason, and act reliably? We cast this as a theory‑ and model‑agnostic challenge where a system must infer trustworthy actions for increasingly distant future states as the observation‑reality gap widens. The effective Speed of Coherent Cognition ($S_C^{\mathrm{eff}}$) is defined as the highest tested Speed of Reality ($S_R$) at which the system continuously meets a predefined behavioural reliability criterion.
As a proof of concept, reinforcement‑learning agents are trained at a baseline $S_R$ and then evaluated without retraining while $S_R$ is increased up to $50\times$, keeping computational resources fixed. We compare Transformer and gated‑MLP baselines with $Co^4$, an architecture inspired by cellular mechanisms implicated in conscious processing and awake thought. Across all scenarios, $Co^4$ achieves a substantially higher $S_C^{\mathrm{eff}}$ than the baselines, supporting $S_C^{\mathrm{eff}}$ as a new benchmark and suggesting that scaling coherent cognition is a principle for reliable AI.
Review