The Prediction Trap: A Philosophical Critique of LeCun's World Model Architecture

Yann LeCun's Joint Embedding Predictive Architecture (JEPA) represents the mostambitious attempt to achieve machine intelligence through world models—internalrepresentations that predict future states in abstract space. This paper presents fourphilosophical critiques demonstrating that this paradigm cannot succeed regardless ofscale or refinement. First, JEPA perpetuates a causal illusion: statistical predictioncannot yield causal understanding because causation is a cognitive label imposedthrough intervention, not a pattern extracted from observation. Second, the predictiveparadigm is frame-bound: it operates within fixed horizons and cannot question its ownpredictive frame—as evidenced by V-JEPA's collapse from 98% accuracy on IntPhys tonear-chance on IntPhys 2. Third, JEPA maintains an inverted architecture that privilegesabstraction over embodied grounding, circumventing rather than solving the symbolgrounding problem. Fourth, world models cannot handle emergence: they extrapolateexisting patterns but fail when confronting genuine novelty. We propose dissipativeintelligence as an alternative—cognition as minimal-energy reactive coupling rather thanprediction. Evolution discovered this solution billions of years ago: organisms survivethrough reflex, not world models. Intelligence is not about predicting the world but aboutstructuring oneself to act without prediction.