The Fractured Entangled Representation Hypothesis (Intro)

This article discusses a critical perspective on current AI, suggesting that its impressive performance is superficial. It introduces the "Fractured Entangled Representation Hypothesis," arguing that current AI's internal understanding is disorganized and lacks true structural coherence, akin to a "total spaghetti." The article contrasts this with a more intuitive and powerful approach, referencing Kenneth Stanley's "Picbreeder" experiment, which generates AI with a deeper, bottom-up understanding of the world. The core argument centers on the difference between memorization and genuine understanding, advocating for methods that prioritize internal model clarity over brute-force training.