A new paper proposes that translation serves as both a foundational concept and a guiding principle for current generative AI models. The author argues that the pursuit of "world models" in AI, which emphasizes internal consistency and grounding, relies on a "statistical ontology." This ontology assumes reality has a computable statistical structure recoverable through data, rather than being shaped by human interpretation. The paper traces this idea back to Warren Weaver's early work on machine translation and contrasts it with Translation Studies' focus on opacity and power dynamics. AI
IMPACT Challenges the fundamental assumptions of current generative AI, suggesting a need to reconsider its relationship with human interpretation and labor.
RANK_REASON Academic paper published on arXiv discussing theoretical underpinnings of generative AI. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- generative artificial intelligence
- Hugging Face
- machine translation
- translation studies
- Warren Weaver
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →