Why Sampling Is Not Choosing: Intentionality, Agency, and Moral Responsibility in Large Language Models

The paper argues against the attribution of agency and moral responsibility to large language models (LLMs), asserting that their outputs stem from probabilistic mappings rather than intrinsic intentionality or self-attributed action. It emphasizes that stochastic sampling in LLMs does not equate to genuine choice or authorship, challenging prevailing notions that these models can exhibit moral reasoning or agency. This analysis is crucial for practitioners to understand the limitations of LLMs in ethical contexts and the implications for their deployment in decision-making systems.