Elamur Research⌕
☰
home / papers / 2023 / Exploring the Jungle of Bias: Political Bias Attribution in Language Models via Dependency Analysis

Exploring the Jungle of Bias: Political Bias Attribution in Language Models via Dependency Analysis

Paper · 2023 · David F. Jenny, Yann Billeter, Mrinmaya Sachan et al.

Summary

LLMs often appear politically biased, but the internal causes of that bias are poorly understood. So are the knock-on effects of debiasing on unrelated behaviour. The authors frame this with causal fairness analysis, using the Standard Fairness Model. A protected attribute (a candidate's party) can shape an outcome (the model's rating) directly, through mediators or via confounders. These intermediate variables are hidden in a black-box LLM, so the authors draw candidate attributes from domain literature. They then prompt ChatGPT to score debate excerpts on them, such as evasiveness, respectfulness and persuasiveness. The scores are combined with contextual metadata like year, topic and party. Activity Dependency Networks (ADNs) then reveal how the attributes depend on each other. In this study, ChatGPT judged argument quality in US presidential debates. Its unequal treatment of the parties is at least partly explained by confounding and mediating attributes and by model misalignment. This suggests that naive fine-tuning to remove bias may silently distort other value associations.

Key points

  • Corpus: Commission on Presidential Debates (CPD) transcripts of presidential and vice-presidential debates since 1960. It covers 47 debates, about 50K sentences and 810K words.
  • Preprocessing: speaker turns were grouped into overlapping slices of 2,500 BPE tokens (≈1,875 words) with 10% overlap.
  • Method: attributes are extracted through prompting and dependencies are estimated non-parametrically with Activity Dependency Networks (Kenett et al., 2012). "Dependency analysis" here means ADNs, not syntactic parsing (no UD, treebank or LAS).
  • Authors: Yann Billeter and David F. Jenny (equal contribution), with Bernhard Schölkopf, Mrinmaya Sachan and Zhijing Jin.
  • Code and data are released at github.com/david-jenny/LLM-Political-Study.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓ETH ZurichZurich · Switzerland84 works · 9,181 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksSafety Research · 626 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksBias, fairness & safety · 587 works

Cite

Cite as David F. Jenny, Yann Billeter, Mrinmaya Sachan, Bernhard Scholkopf and Zhijing Jin. 2023. Exploring the Jungle of Bias: Political Bias Attribution in Language Models via Dependency Analysis. arXiv:2311.08605.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2311.08605 · arXiv 2311.08605

@misc{jenny2023exploring,
  title = {{Exploring the Jungle of Bias: Political Bias Attribution in Language Models via Dependency Analysis}},
  author = {David F. Jenny, Yann Billeter, Mrinmaya Sachan, Bernhard Scholkopf and Zhijing Jin},
  year = {2023},
  doi = {10.48550/arxiv.2311.08605},
  eprint = {2311.08605},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2311.08605}
}

References

26 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abubakar Abid, Maheen Farooqi, and James Zou. 2021. Persistent anti-muslim bias in large language models. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society. ACM. 1
  2. Rohan Anil, Andrew M. Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H. Clark, Laurent El Shafey, Yanping Huang, Kathy Meier-Hellstern, Gaurav Mishra, Erica Moreira, Mark Omernick, Kevin Robinson, Se… — PaLM 2 Technical Report
  3. Su Lin Blodgett, Gilsinia Lopez, Alexandra Olteanu, Robert Sim, and Hanna Wallach. 2021. Stereotyping Norwegian salmon: An inventory of pitfalls in fairness benchmark datasets. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint… find
  4. Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc. 1
  5. Sunipa Dev, Emily Sheng, Jieyu Zhao, Aubrie Amstutz, Jiao Sun, Yu Hou, Mattie Sanseverino, Jiin Kim, Akihiro Nishi, Nanyun Peng, and Kai-Wei Chang. 2022. On measures of biases and harms in NLP. In Findings of the Association for Computational Linguistics: AACL-IJCNLP 2022, pages 246–267, Online only… find
  6. Shangbin Feng, Chan Young Park, Yuhan Liu, and Yulia Tsvetkov. 2023. From pretraining data to language models to downstream tasks: Tracking the trails of political biases leading to unfair NLP models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume…
  7. Hila Gonen and Yoav Goldberg. 2019. Lipstick on a pig:. In Proceedings of the 2019 Conference of the North. Association for Computational Linguistics. 1 find
  8. Jochen Hartmann, Jasper Schwenzow, and Maximilian Witte. 2023. The political ideology of conversational AI: Converging evidence on ChatGPT’s pro-environmental, left-libertarian orientation. SSRN Electronic Journal. 1
  9. Dror Y. Kenett, Tobias Preis, Gitit Gur-Gershgoren, and Eshel Ben-Jacob. 2012. Dependency Network and Node Influence: Application to the study of financial markets. International Journal of Bifurcation and Chaos, 22(07):1250181. 2, 5 find
  10. Katelyn X. Mei, Sonia Fereidooni, and Aylin Caliskan. 2023. Bias against 93 stigmatized groups in masked language models and downstream sentiment classification tasks. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency. 1 find
  11. Pranav Narayanan Venkit, Sanjana Gautam, Ruchi Panchanadikar, Ting-Hao Huang, and Shomir Wilson. 2023. Nationality bias in text generation. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 116– 122, Dubrovnik, Croatia. Association… find
  12. OpenAI. 2023. Gpt-4 technical report. 1
  13. Davor Petreski and Ibrahim C. Hashim. 2022. Word embeddings are biased. but whose bias are they reflecting? AI & SOCIETY, 38(2):975–982. 1 find
  14. Drago Plecko and Elias Bareinboim. 2022. Causal fairness analysis. 2 find
  15. Machel Reid, Nikolay Savinov, Denis Teplyashin, Dmitry Lepikhin, Timothy Lillicrap, Jean baptiste Alayrac, Radu Soricut, Angeliki Lazaridou, Orhan Firat, Julian Schrittwieser, Ioannis Antonoglou, Rohan Anil, Sebastian Borgeaud, Andrew Dai, Katie Millican, Ethan Dyer, Mia Glaese, Thibault Sottiaux, B… — Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context
  16. Adi Robertson. 2024. Google apologizes for ‘missing the mark’ after gemini generated racially diverse nazis. 2 find
  17. Salvatore Ruggieri, Jose M. Alvarez, Andrea Pugnana, Laura State, and Franco Turini. 2023. Can we trust fair-ai? Proceedings of the AAAI Conference on Artificial Intelligence, 37(13):15421–15430. 2 find
  18. Juliana Shihadeh, Margareta Ackerman, Ashley Troske, Nicole Lawson, and Edith Gonzalez. 2022. Brilliance bias in GPT-3. In 2022 IEEE Global Humanitarian Technology Conference (GHTC). IEEE. 1 find
  19. Marco R Steenbergen, André Bächtiger, Markus Spörndli, and Jürg Steiner. 2003. Measuring political deliberation: A discourse quality index. Comparative European Politics, 1(1):21–48. 3, 4 find
  20. Zeerak Talat, Aurélie Névéol, Stella Biderman, Miruna Clinciu, Manan Dey, Shayne Longpre, Sasha Luccioni, Maraim Masoud, Margaret Mitchell, Dragomir Radev, Shanya Sharma, Arjun Subramonian, Jaesung Tae, Samson Tan, Deepak Tunuguntla, and Oskar Van Der Wal. 2022. You reap what you sow: On the challen…
  21. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  22. Oskar van der Wal, Dominik Bachmann, Alina Leidinger, Leendert van Maanen, Willem Zuidema, and Katrin Schulz. 2022. Undesirable biases in nlp: Averting a crisis of measurement. 1 find
  23. Eva Maria Vecchi, Neele Falk, Iman Jundi, and Gabriella Lapesa. 2021. Towards argument mining for social good: A survey. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Lo… find
  24. Pranav Narayanan Venkit, Sanjana Gautam, Ruchi Panchanadikar, Ting-Hao Huang, and Shomir Wilson. 2023. Unmasking nationality bias: A study of human perception of nationalities in AI-generated articles. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. ACM. 1 find
  25. Henning Wachsmuth, Nona Naderi, Yufang Hou, Yonatan Bilu, Vinodkumar Prabhakaran, Tim Alberd-ingk Thijm, Graeme Hirst, and Benno Stein. 2017. Computational argumentation quality assessment in natural language. In Proceedings of the 15th Conference of the European Chapter of the Association for Compu…
  26. Junzhe Zhang and Elias Bareinboim. 2018. Fairness in decision-making — the causal explanation formula. In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence and Thirtieth Innovative Applications of Artificial Intelligence Conference and Eighth AAAI Symposium on Educational A… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jenny-2023-exploring-jungle-bias-political-bias/