Upstream Dual-Use: Modelling, Epistemic Authority, and Institutional Harm in AI Systems

Valeriano Grasso

Track: tiny / short paper (up to 4 pages)

Keywords: Dual-use AI, Epistemic authority, AI accountability, Science and Technology Studies, Machine learning ethics

TL;DR: Dual-use in AI does not begin at deployment, but at the moment models are trained and legitimized as epistemic authorities.

Abstract:

The dual-use problem of artificial intelligence is often treated as arising downstream issue arising at
the point of deployment. We explore the hypothesis that dual-use is rooted upstream, in modelling
and training practices that confer epistemic authority on AI systems before use. Focusing on design
choices, data selection, and validation processes, we show how institutional reliance on AI systems
displaces responsibility from human and organizational collectives onto the models themselves. With
brief reference to cases, the paper reframes dual-use as a problem of epistemic legitimation, with
direct relevance for accountability and peace-oriented AI research.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 4