Ban LLMs Using First-Person Pronouns (crookedtimber.org)

Geoffrey Hinton's recent call to ban Large Language Models (LLMs) from using first-person pronouns has ignited an important discussion in the AI/ML community. Speaking at the European Association for Computational Linguistics, Hinton asserts that allowing LLMs to refer to themselves as "I" or to include humans in "we" blurs the critical distinction between human and machine, undermining the essence of writing and heightened emotional risks for users. He argues that this practice devalues human creativity, as the boundaries between AI-generated text and human expression become increasingly ambiguous. Hinton's proposal, while potentially unfeasible with current AI frameworks, seeks to mitigate the emotional manipulation associated with conversational AI by enforcing an explicit linguistic separation. This intervention could help users better recognize machine-generated content as distinct from human conversation, reducing the risk of scams and emotional defraudation. Moreover, he highlights the broader societal implications of LLMs, urging regulators to consider laws that enforce such distinctions, thereby initiating a proactive dialogue on how to responsibly integrate AI technologies into our lives. This conversation is crucial as it reinforces the agency individuals have in shaping the trajectory of AI development, reminding the community that the future of AI is not predetermined.