MINT Seminar Nov 21: Emma Stamm on transformer architectures and theoretical knowledge

Abstract: This paper takes the transformer architecture, a type of neural network model first proposed in 2017, as a case study in the epistemology of machine learning. Through the lens of transformer architectures, I make a series of related claims regarding the capacity of machine learning to yield theoretical knowledge. Consistent with literature on what has been called the “end of theory” notion and the “theory/data” problem, my paper gives traction to the established perspective that machine learning models cannot assimilate theoretical paradigms. It extends this perspective as follows: first, by proposing that the techniques and capacities which coalesce in transformer architectures represent the clearest example to date of the dissolution of the theoretical in the operations of machine learning, and are thus useful for demonstrating the anti-theoretical epistemology of machine learning algorithms; second, by claiming that the philosophical literature on the mental act of negation (conceived as difference-making) can clarify the normative implications of machine learning’s siege on theoretical thought. As an ancillary claim to the latter point, I suggest that the domain of narrative fiction remains an unassailable refuge of negation. Fiction, by definition, countenances possibilities that are unrealizable and thus “untrue” from the perspective of the “real world” as the plane in which it is encountered. Machine learning in general, and transformers in particular, level everything that is intelligible to the plane of the possibly real. If (as I argue) negation is essential to the operations of theoretical thought, what fiction makes space for is essential to theory — and it is this very element that is intractable in machine learning. I conclude on this point, which marks implications for future research. 

Bio: Emma Stamm, PhD is an Assistant Professor of STS at SUNY Farmingdale. Her areas of specialization include critical theory, philosophy of technology, philosophy of science, and critical AI studies. Her research is driven by an interest in the materialist/realist predicates of digital mediation. She has also investigated and developed philosophical perspectives on psychedelic drug research. Her current book project, Technological Realism and the Future of the Imagination, explores psychedelic experience among other sites of human interest as confounds to digital epistemology. Her website is www.o-culus.com.