Today's AI vision systems have memorized an astonishing range of information and patterns.  When given infinite computing power, the danger of AI is that it can override logic.   For many ordinary tasks, this logic override is rare, since models are trained on vast amounts of data and have effectively memorized the answers. However, when faced with an unfamiliar or complex task, this pattern-checking style of learning turns against them, and AI resorts to hallucination, while humans can reason on the fly: resample the environment to gather information, efficiently balancing between checking and looking (searching).