AI's Carwash Test: Do Statistical Shortcuts Override Logical Reasoning?

In a revealing experiment dubbed the 'Carwash Test,' AI researcher Christopher Horrocks probes a fundamental weakness in modern artificial intelligence: the tendency to rely on surface-level statistical patterns rather than true logical reasoning. The test presents a simple problem designed to trip up systems that prioritize common word associations over careful object tracking. Over 27 runs across 12 different AI systems, the results highlight a persistent challenge in the field of virtual intelligence. The full essay, including a results matrix, is available on Horrocks' Substack, and he discusses the implications in the latest episode of his podcast, 'Virtual Intelligence in Action.'