Research

Artificial Intelligence, Scientific Progress, and the Maker’s Knowledge Tradition

The center of this direction is a reinterpretation of the maker’s knowledge tradition’s core intuition: that we know best what we’ve made. From Bacon through Hobbes and Vico, philosophers have articulated makers’ epistemic privilege through the concept of authorship: the authorship relation grants makers transparent access to the objects of their knowledge. In “Maker’s Knowledge: A Material Severity Account” (under review) I argue that authorship is the wrong ground. Making is epistemically generative because making is discriminatingly unforgiving: the process of trying and failing reveals to the maker where they lack relevant knowledge.

“Scientific Works and the Transfer of Technique” explains how experimental techniques transfer between scientific domains that don’t share a theory. I argue that the concept of scientific works, developed to describe how science makes progress, explains when such transfer succeeds and when it fails better than extant alternatives. It is the first article-length treatment of the concept that powers the operative account of scientific progress, which grows out of my dissertation, Engineering Progress in Science (University of Cincinnati, 2025).

“Maker’s Knowledge in Machine Learning” applies the material severity account to an apparent puzzle in the philosophy of artificial intelligence: AIs like ChatGPT are among the most impressive artifacts humans have ever made, yet their makers seemingly lack much knowledge about how they work. The puzzle is dissolved when the authorship reading is replaced by the severity reading, which predicts the distribution of knowledge we observe: engineers fluent in training techniques and architectures but not in why particular outputs arise.

Because this direction concerns making minds, my method matches it. In “When Does the Body Matter?” (forthcoming, SAB 2026, Springer LNAI) I evolve artificial agents and measure what becomes of their neural dynamics when their coupling to the environment is severed, and my engineering master’s thesis built sensors modeled on the whiskers animals use to feel their way in the dark. With Moti Mizrahi I am testing the progress debate against evidence, coding the complete run of Nobel lectures and Nobel Committee ceremony speeches, 1901–2025, 1,553 documents in all, against the four leading philosophical accounts of what scientific progress consists in.

Engineering, Artificial Intelligence, and the Common Good

The second direction unpacks what AI engineers owe everyone else. In Lucken and Feiten (2025, Studies in History and Philosophy of Science), my co-author and I connect radical embodied cognitive science to public engagement with science. The two fields reject parallel mistakes, the deficit model in science communication and the information-processing paradigm in cognitive science, and the enactivist concept of participatory sense-making explains why active, dialogical engagement succeeds where one-way transmission fails. We argue that democratizing AI means restructuring what affected communities can do with these systems and who participates in shaping them. A Guide for Academic Researchers Conducting Science Outreach (under contract with Cambridge University Press) turns that argument into concrete guidance for designing, running, and evaluating scientific outreach.

“An Identity-Based Ethics for Engineering” (forthcoming, Science and Engineering Ethics) argues that engineering is a morally significant professional identity: a life-shaping practice whose regulative ideal is the development of central human capabilities. What engineers owe everyone else derives from what it means to be an engineer. A new project applies the account to the engineers building frontier AI systems at labs such as Anthropic, OpenAI, and Google DeepMind: if engineering’s regulative ideal is the development of human capabilities, then frontier AI engineering answers for whether its systems develop or degrade the capabilities of the people they touch, and professional identity is the right locus of that obligation.

With Ezgi Yildiz and Umut Yüksel, I am studying how the European Court of Human Rights has treated non-refoulement claims as member states press for a more restrained court. We built a large language model classifier over 14,196 Article 3 judgments (1967–2025), validated it against 2,884 hand-coded cases, and ran trend analyses; our first results appeared in EJIL: Talk! in June 2026.