ESHS/HSS Annual Meeting

State Knowledge and the Performance of (Dis)ambiguity

• In Event: CTRL ALT SHIFT. Expériences, Knowledge, Activism, Methods - 3

English Abstract

The rise of statistical reasoning in the seventeenth century introduced novel tools for representing the limits of knowledge about the world: Probability was expressed through the language of odds and distributions; and uncertainty was quantified in the form of confidence intervals and standard errors. When modernizing nation-states began to "see" through government statistics, those tools became integral to the exercise of bureaucratic power and to the government of large populations and complex economies. They remained so even as technologies evolved and still underpinned the development of predictive risk models and the ordinalization of state knowledge in the early twenty-first century. However, the current generation of AI chain-of-thought models departs from this approach by foregrounding the performance of (dis)ambiguity. Such models—which are rapidly entering the domain of public administration—aim to mimic human reasoning by iterating through multiple responses while making implicit disambiguation choices. Ambiguity is performed internally but is often invisible to end-users or subject to post-hoc rationalization. This raises important questions about the nature of state knowledge in the age of AI: How do state officials engage with technologies that perform (dis)ambiguity rather than quantifying uncertainty? What are the implications for bureaucratic accountability? And who polices the limits of state knowledge?

Author

• Martin Eiermann, UW-Madison