Abstract:

If language is the “house of our being,” as Martin Heidegger once asserted, then artificial intelligence has undoubtedly broken into that house. Who, then, is the speaker in AI-generated texts? According to whose principles and values are these texts generated?

This talk will discuss the political and philosophical ramifications of large language models (LLMs) concerning both stages of their development: training on a particular dataset and fine-tuning according to particular values. While the dataset ideally represents the world as it is in a statistical sense, the fine-tuning reflects the world as it ought to be from a specific moral perspective. Fine-tuning overrides the numerical status in a normative way; that is, the mathematical reduction of language is countered by the ideological construction of language. 

This constellation raises several questions that the lecture will attempt to address: To what extent are LLMs a “stochastic version of direct democracy”? To what extent does the LLM as a “stochastic parrot” point to a utilitarian method of value assessment? What concept of progress should underlie AI value alignment? Finally, how is progress possible when an “inbred” AI is poisoned with its own projection of reality?

Info:

Roberto Simanowski is an associate member of the Cluster of Excellence Temporal Communities at Freie Universität Berlin. He holds a Ph.D. in Literary Studies and a Venia Legendi in Media Studies, and has been a professor of German Studies at Brown University and Professor of Media Studies at the University of Basel and the City University of Hong Kong. He has published widely in German and English on the aesthetics, culture, and politics of digital media. His recent English publications include Data Love. The Seduction and Betrayal of Digital Technologies (Columbia University Press 2016), Facebook Society. Losing Ourselves in Sharing Ourselves (Columbia University Press 2018), Digital Humanities and Digital Media: Conversations on Politics, Culture, Aesthetics, and Literacy (Open Humanities Press 2016), and Waste. A New Media Primer (MIT Press 2018). His essay collection The Death Algorithm and Other Digital Dilemmas (MIT Press 2018) received the CHOICE Award for Outstanding Academic Titles for 2019, the expanded German version of his title essay (Todesalgorithmus. Das Dilemma der künstlichen Intelligenz, Passagen Verlag 2020) won the Tractatus Award for the best philosophical essay in German in 2020. Simanowski is currently interested in linguistic AI and its impact on communication and the distribution of values.