Artificial epistemology
When do belief, evidence, and revision become legitimate explanatory concepts for AI? I connect formal and naturalized epistemology with testable architectural conjectures.
Belief revision · Formal logic · Epistemic agency
Artificial intelligence · Philosophy · Research engineering
I study how evidence, intervention, and architecture shape AI behavior, and what would make beliefs and reasons useful explanations of artificial agents.
University of Maryland, ARLIS
University of Kansas, Center for Cyber-Social Dynamics
01 Research
When do belief, evidence, and revision become legitimate explanatory concepts for AI? I connect formal and naturalized epistemology with testable architectural conjectures.
Belief revision · Formal logic · Epistemic agency
How should agents maintain knowledge, use tools, and coordinate over time? My work explores modular agent architectures, evidence-grounded research, and multi-agent systems.
Agent architecture · Knowledge systems · Bounded autonomy
What does an evaluation actually establish? I work on operational testing, adversarial evaluation, and model provenance, with attention to controls, evidence, and the limits of inference.
Operational evaluation · Model lineage · Adversarial testing
02 Selected work
Public software and research.
Institutional work is identified separately.
A modular multi-agent harness with an evidence-grounded research suite. The knowledge layer treats claims and their support as first-class objects, so changes in evidence can propagate through a web of belief.
In development · Kernel and knowledge-layer redesign
An open-source signal-processing and machine-learning toolkit. As lead developer from 2023 to 2025, I worked on dataset generation, transforms, spectrogram tools, training workflows, and GNU Radio integration.
RF machine learning · PyTorch · Signal processing
A research implementation of a whole-response latent predictor, combining a JEPA-style architecture, mixture-of-experts routing, and episodic memory. Executable synthetic worlds provide the correctness signal.
Experimental implementation · Not a validated architecture
03 Writing
Refereed publications
Justin P. Mullins and John Symons
36(3), exag018 · DOI: 10.1093/logcom/exag018
E. Oh, J. Mullins, M. Carrick, M. Vondal, J. Hoffman, F. Leonardo, P. Toliver, and R. Miller
P. Vallance, E. Oh, J. Mullins, M. Gulati, J. Hoffman, and M. Carrick
9(1)
Work in progress
Toward an Engineering Science of Epistemic Agency
Developing testable architectural conjectures from questions about belief, evidence, revision, communication, and action.
V. Chang, J. Mullins, and A. Walter-Higgins
04 About
I am a Visiting Associate Research Scientist at ARLIS, University of Maryland, and a Research Fellow in AI/ML and NLP at the University of Kansas’s Center for Cyber-Social Dynamics.
My background spans formal logic, philosophy of science, RF machine learning, and AI evaluation. Earlier work in the U.S. Intelligence Community informs my interest in how systems behave under realistic operational conditions.
I earned my Ph.D. in Philosophy at the University of Kansas in 2023. My dissertation, Quine’s Methodology: Science, Logic, and Ontology, examined the relationship between scientific inquiry, logic, and ontological commitment.