Mutual Prediction

Coevolution reconceived as reciprocal predictive ability: species that better predict each other gain fitness, and "death … or failure to reproduce" is the ultimate prediction error. Loewith and Street sort predictive ability into four levels — model-free (genetic/phenotypic, e.g. bacteria reading host-plant signals), model-based (an updatable world model), social-model-based (theory of mind), and a speculative "complete model" — and argue that asymmetry across those levels, not raw capability, determines whether a relationship is mutualist, commensal, parasitic or amensalist, and therefore where power sits. Applied to human–AI relations it maps ELIZA, frontier LLMs, an "AI neuroscientist," unnoticed social modelers, ASI, and cyborg fusion (where prediction ends because the boundary does).