Toward Non-Human-Centric AI: Conceptual Note on the Limitations of Language-Based Models and the Possibility of Ecological Intelligence

This conceptual note proposes an original perspective on the limitations of current large language models (LLMs) and the potential emergence of non-human-centric AI. Modern LLMs are trained exclusively on human language and therefore cannot model, approximate, or even hallucinate from non-human behavioral or ecological data. This document identifies a conceptual gap between AI research and fields such as animal behavior, plant signaling, ecology, and Earth system science. It hypothesizes that AI trained on non-human data—such as movement patterns, physiological responses, or ecosystem feedback loops—could develop fundamentally different reasoning structures beyond human linguistic logic. The note outlines the implications of such models and argues that exploring this direction may expand our understanding of intelligence itself.