AI News

The paper proposes a theory- and model-agnostic way to study when an intelligent system can no longer reliably perceive, reason, and act as reality changes faster. It defines effective Speed of Coherent Cognition (S_C^eff) as the highest tested Speed of Reality (S_R) at which a system continuously meets a predefined behavioral reliability criterion from a baseline. As a proof of concept, reinforcement-learning agents are trained at the baseline environment speed and then evaluated without retraining while S_R increases by as much as 50 times, with computational resources held fixed. The experiments compare Transformer and gated-MLP baselines with Co^4, an architecture inspired by cellular mechanisms associated with conscious processing and awake thought. Across the tested scenarios, Co^4 achieves substantially higher S_C^eff than the baseline architectures. The authors present this result as motivation for using S_C^eff as a new benchmark and for treating the scaling of coherent cognition as a principle for reliable AI. The abstract does not establish how broadly the result generalizes beyond the tested scenarios or architecture comparisons.