On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
Bender et al., 2021, On the Dangers of Stochastic Parrots (Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21))
Subscriber deck — free during early access
1 cards in “4.3 Encoding Bias” · whole deck
4.3 Encoding Bias
- Why can’t bias audits of LMs find all risks, according to Bender et al.?Audits rely on automated measures (sentiment, toxicity) that are themselves biased (e.g. toxicity scores rise with identity terms); they need a priori knowledge of which social categories matter, which is culture-bound; and certifying “safety” requires engaging with the systems of power behind harms, which is inherently political.
- Bender et al., 2021, On the Dangers of Stochastic Parrots (Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21)), §4.3, pp. 614–615
Source
- Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Shmargaret Shmitchell. 2021. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21), pp. 610–623. License: CC BY 4.0. Flashcards adapted from this work. doi:10.1145/3442188.3445922