Emily Bender Debunks Misconceptions About 'Stochastic Parrots'
OpenAI
Google/DeepMind
Emily M. Bender, co-author of the landmark 2021 paper 'On the Dangers of Stochastic Parrots,' clarifies common misunderstandings about the metaphor, arguing that it specifically describes large language models (LLMs) producing synthetic text, not all AI. She also discusses the vagueness of the term 'artificial intelligence,' the exploitative labor practices in AI, and the importance of recognizing human sense-making in evaluating language models.
In a recent interview with IEEE Spectrum, University of Washington computational linguistics professor Emily M. Bender addressed misconceptions about the 'stochastic parrots' metaphor from her 2021 paper. She emphasized that the term applies only to large language models generating synthetic text through statistical prediction, not to other AI systems like chess engines or AlphaFold. Bender criticized the broad use of 'artificial intelligence' as obscuring technological differences and aiding corporate hype. She noted that the paper's section on synthetic text became more relevant after OpenAI launched ChatGPT. Bender regretted not covering exploitative labor practices and data theft in the original paper. She also highlighted that when LLM outputs make sense, it is due to human interpretation, not machine understanding.
- Abbreviations
- LLM = Large Language Model — большая языковая модель
Source: IEEE Spectrum AI —
original