Authorship and the Politics and Ethics of LLM Watermarks
Abstract: Recently, watermarking schemes for LLMs have been proposed to distinguish text generated by machines and by humans. The present paper explores philosophical, political, and ethical ramifications of implementing and using watermarking schemes. A definition of authorship that includes both machines (LLMs) and humans is proposed to serve as a backdrop. It is argued that private watermarks may provide private companies with sweeping rights to determine authorship, which is incompatible with traditional standards of authorship determination. Then, possible ramifications of the so-called entropy dependence of watermarking mechanisms are explored. It is argued that entropy may vary for different, socially salient groups. This could lead to group dependent rates at which machine generated text is detected. Specifically, groups more interested in low entropy text may face the challenge that it is harder to detect machine generated text that is of interest to them.
- Aaronson, S. 2023. Watermarking of LLMs. Slides, Simons Institute, Berkeley, August 17, 2023.
- The Stable Entropy Hypothesis and Entropy-Aware Decoding: An Analysis and Algorithm for Robust Natural Language Generation. arXiv preprint arXiv:2302.06784 .
- Undetectable Watermarks for Language Models. arXiv preprint arXiv:2306.09194 .
- Coeckelbergh, M. and D. J. Gunkel. 2023. ChatGPT: deconstructing the debate and moving it forward. AI & SOCIETY : 1–11.
- Cover, T. M. and J. A. Thomas. 2006. Elements of Information Theory. Wiley, second ed.
- de Haas, H. 2023. How Migration Really Works. Penguin.
- Eckert, S. 2015. The Guttenberg plagiarism scandal: myths through Germany’s leading news magazines. Journal of Communication Inquiry 39(3): 249–272.
- Ehret, K. 2016. An information-theoretic approach to language complexity: variation in naturalistic corpora. Ph.D. thesis, Dissertation, Albert-Ludwigs-Universität Freiburg, 2016.
- Publicly detectable watermarking for language models. arXiv preprint arXiv:2310.18491 .
- Foucault, M. 1998. What is an Author? In J. D. Faubion, ed., Aesthetics, Method, and Epistemology, vol. 2. New York: The New Press, pp. 205–222.
- An ugly truth: Inside Facebook’s battle for domination. Hachette UK.
- The ethical need for watermarks in machine-generated language. arXiv preprint arXiv:2209.03118 .
- Isaak, J. and M. J. Hanna. 2018. User data privacy: Facebook, Cambridge Analytica, and privacy protection. Computer 51(8): 56–59.
- A watermark for large language models. arXiv preprint arXiv:2301.10226 .
- Kontoyiannis, I. 1997. The complexity and entropy of literary styles. Tech. rep., Department of Statistics, Stanford University Stanford, CA, USA.
- Languages with more speakers tend to be harder to (machine-) learn. Scientific Reports 13(1): 18521.
- Robust distortion-free watermarks for language models. arXiv preprint arXiv:2307.15593 .
- MacKay, D. 2003. Information Theory, Inference, and Learning Algorithms. Cambridge University Press.
- Generative AI entails a credit–blame asymmetry. Nature Machine Intelligence : 1–4.
- Guttenberg soll bei Doktorarbeit abgeschrieben haben. Süddeutsche Zeitung.
- ChatGPT vs Human-authored Text: Insights into Controllable Text Summarization and Sentence Style Transfer. arXiv preprint arXiv:2306.07799 .
- Robust speech recognition via large-scale weak supervision. In International Conference on Machine Learning. PMLR, pp. 28492–28518.
- Rodriguez, M. C. 2010. History of supersymmetric extensions of the Standard Model. International Journal of Modern Physics A 25(06): 1091–1121.
- Authorship and ChatGPT: a Conservative View. Philosophy & Technology 37(1): 34.
- Ethical and social risks of harm from language models. arXiv preprint arXiv:2112.04359 .
- Watermarks in the sand: Impossibility of strong watermarking for generative models. arXiv preprint arXiv:2311.04378 .
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Collections
Sign up for free to add this paper to one or more collections.
