Year: 2026
Type: conference-paper
Abstract: I argue that the practice of applying generic deep neural network (DNN) architectures with minimal theoretical constraints falls short of the conditions required for scientific representation on both substantive or deflationary accounts. Substantive views fail because the characteristic interpretive activities that establish representation relations are absent from generic DNN practice. Deflationary views fare better but risk trivializing the concept of surrogative inference when applied to generic DNNs. I then propose that theoretically-motivated architectural constraints function as a form of indirect characterization that grounds deflationary scientific representation.
Author: Kieval, Phillip H.
Institution: University of Florida
Language: English
Subfield: History and Philosophy of Science
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7164795065
