Understanding ''Democratization'' in NLP Research
Abstract: Despite increasing discussion of the ''democratization'' of natural language processing and machine learning research, the use of this term and its connections to democracy have not been thoroughly studied.
Given the rich history of democracy, understanding what AI researchers mean by ''democratization'' is important for ensuring that we are accurately representing public participation in and control of the field.
Thus, we conduct a large-scale, mixed-methods analysis of every use of democracy-related terms among all papers published in the ACL Anthology or at ICLR, ICML, or NeurIPS ($N = 507$ papers); we do this to uncover the themes, values, and concepts that researchers associate with democracy.
In addition, we examine how deeply papers that mention democracy engage with the concept via their text and citations.
Ultimately, we find that ''democratization'' mostly signals broadening access or use of technologies, especially without expertise.
In contrast, researchers' conceptualizations of democracy are diverse and grounded in theories of deliberation and debate.
Moreover, we observe that papers that mention democracy often do not meaningfully treat democracy or draw on democratic theories from outside NLP.
Based on our findings, we urge responsible use of the term ''democratization'' and greater engagement with theories of democracy towards enriching our discussions of AI access and governance.
Paper Type: Long
Research Area: Ethics, Bias, and Fairness
Research Area Keywords: ethical considerations in NLP applications,reflections and critiques,values and culture
Contribution Types: Data analysis, Surveys
Languages Studied: English
Submission Number: 791
Loading