This position paper argues that humans possess an evolved instruction-following bias, an innate inductive bias shaped by evolution to interpret and execute linguistic instructions. This cognitive feature enables rapid instructed task learning (RITL) and allows for the fast generalization of behavior from language.
The authors posit that this human bias functions analogously to how large language models leverage instruction tuning to achieve zero-shot task performance. While AI instruction-following currently relies on specialized training protocols, the paper suggests it arises as an innate feature in humans.
The study synthesizes evidence from cognitive science, neuroscience, and machine learning research to support this hypothesis. It outlines testable predictions and calls for more interdisciplinary research to investigate instruction-following as a unifying mechanism for rapid task learning.