If you think, movements like the Social Hygiene Movement or the Temperance Movement are human follies of the past, you might be very mistaken.
I just happen to study following preprint research paper in computational linguistics: Language Models are Few-Shot Learners
The authors of this paper are affiliated with OpenAI. Several of the researchers, who authored this paper, are highly cited and regarded.
The paper is about one of the very largest language models ever built to date, GPT-3 with about 175 billion parameters. The paper is unusually long with 74 pages written by 30 authors.
Well, the authors did not fail to address issues like social responsibility or social implications in this paper.
In their chapter about e.g. gender bias (pages 36-37), they discover in all earnest that there is a distinct bias in the occupational choices of women and men. Oh really! This chapter is quite amusing unless you are indoctrinated by feminist orthodoxy!
Among other things, the authors recommend:
"Ultimately, it is important not just to characterize biases in language systems but to intervene. ... There is room for more research that engages with the literature outside NLP [Natural Language Processing], better articulates normative statements about harm, and engages with the lived experience of communities affected by NLP systems" (page 39). By that they mean perceived and exaggerated gender, racial bias, and religious bias and other biases. Such politically correct obsessions, I am afraid, are all to typical for academics living in a ivory tower bubble!
No comments:
Post a Comment