Censorship of Online Encyclopedias

Abstract

While artificial intelligence provides the backbone for many tools people use around the world, recent work has brought to attention that the algorithms powering AI are not free of politics, stereotypes, and bias. While most work in this area has focused on the ways in which AI can exacerbate existing inequalities and discrimination, very little work has studied how governments actively shape training data. We describe how censorship has affected the development of Wikipedia corpuses, text data which are regularly used for pre-trained inputs into NLP algorithms. We show that word embeddings trained on Baidu Baike, an online Chinese encyclopedia, have very different associations between adjectives and a range of concepts about democracy, freedom, collective action, equality, and people and historical events in China than its regularly blocked but uncensored counterpart -Chinese language Wikipedia. We examine the implications of these discrepancies by studying their use in downstream AI applications. Our paper shows how government repression, censorship, and self-censorship may impact training data and the applications that draw from them. CCS CONCEPTS• Computing methodologies → Supervised learning by classification; • Information systems → Content analysis and feature selection; • Social and professional topics → Political speech.

Ethical and social risks of harm from Language Models

“…Increased surveillance or censorship may amplify existing feedback loops such as "chilling effects", whereby the anticipation of surveillance leads individuals to self-censor (Kwon et al, 2015). In a distinct feedback loop, censorship of web text, for example of online encyclopedias, can then affect the quality of a LM trained on such data (Yang and Roberts, 2021).…”

The Chengdu Biennale and Wikipedia Art Information

“…In attempting to update Wikipedia pages from China, it was discovered that the VPN IP address was blocked by Wikipedia for editing, probably due to previous bad editing by others on the same IP address (see Figure 3). Thus, a combination of the Great Firewall of China due to Chinese censorship (Harrison 2019;Yang & Roberts 2021) and IP address blocking makes both reading and writing of pages problematic. Reading pages can be overcome using VPN access but writing or updating pages proved more difficult.…”

Tech Worker Organizing for Power and Accountability

“…Chouldechova et al audit an algorithm-assisted child maltreatment hotline screening system and identify many of the challenges in implementing such an investigation in practice [16]. Yang et al demonstrate how political censorship of Wikipedia can affect pre-trained models used for general domain NLP algorithms [73]. Bender and Gebru et al critically examine the environmental and financial costs first of large language models and offer some recommendations for curating and documenting datasets more carefully [7].…”