A predictive human model of language challenges traditional views in linguistics and pretrained transformer research
Language and Semiotic Studies January 2, 2025 Torres-Martínez Sergio 7 citations
Language is a cognitive tool that evolved to optimize biological fitness by allowing humans to reconstruct reality through memory and adaptation to uncertainty, reaffirming the Self-as-symbol. Pretrained language models like ChatGPT lack embodied grounding and conscious states, preventing them from adequately modeling the world through language. Human representation cannot be reduced to data structures; transformers are posthuman agents, while humans are purposeful biological agents capable of adjustment and optimization. The capacity to integrate information does not equal phenomenal consciousness, as Information Integration Theory claims. Language models, despite superior computation, lack real consciousness and multiscalar physical experience. The paper anticipates future in silico conceptualizers that may define themselves as phenomenal agents with symbolic contours and goals.